claude-code-plugin-dev

内容来源：README.md（说明文档） · 原始地址 · 查看安装指南

原始内容

# claude-code-plugin-dev [![Build](https://github.com/Cloud-Officer/claude-code-plugin-dev/actions/workflows/build.yml/badge.svg)](https://github.com/Cloud-Officer/claude-code-plugin-dev/actions/workflows/build.yml)

Claude Code plugin for development workflow automation.

## Table of Contents

* [Introduction](#introduction)
* [Installation](#installation)
  * [Quick Start](#quick-start)
  * [Configure Remote MCPs (optional)](#configure-remote-mcps-optional)
  * [Google Workspace MCP (optional)](#google-workspace-mcp-optional)
  * [Complementary Plugins (optional)](#complementary-plugins-optional)
  * [Configure Environment Variables](#configure-environment-variables)
  * [Multi-Account Setups (direnv)](#multi-account-setups-direnv)
  * [Recommended Permissions](#recommended-permissions)
* [Usage](#usage)
  * [Skills](#skills)
  * [Language Servers (LSPs)](#language-servers-lsps)
  * [Local Development](#local-development)
* [Contributing](#contributing)

## Introduction

This plugin provides development workflow automation for Claude Code.

### Features

* Issue tracking integration (GitHub Issues and Jira)
* Automated deep code reviews using parallel agents
* PR generation with commit messages
* Linting and code quality checks
* Automated test generation with coverage enforcement across 10 languages (RSpec, PHPUnit/Pest, pytest, Jest/Vitest, xUnit, Swift Testing/XCUITest, Kotest, Go, GoogleTest, Rust)
* Codebase migration between languages and frameworks via a six-step rulebook/translate/compile/test/verify engine
* Security threat modeling (STRIDE) and code-ownership / bus-factor analysis, ISO 27001-aligned
* Database schema documentation and natural language querying
* Project documentation review (README, architecture, user guide)
* User-facing copy audits (microcopy, voice & tone, i18n readiness)
* SEO / GEO and front-end web quality audits (performance, Core Web Vitals, accessibility)
* Sprint work summaries grouped by repository
* Weekly development and project-status reports from Jira, GitHub, and monday.com
* Tracker coverage checks — verify a planning doc's items exist in monday.com and/or Jira
* Translation asset management via Loco (localise.biz) API
* monday.com board, item, and work-management operations
* Firebase Crashlytics crash analysis via BigQuery
* Figma design-to-code review for Android, iOS, and web
* PayPal payment management
* AWS infrastructure management and documentation search
* Google Cloud infrastructure management via gcloud
* New Relic observability data and alert management
* Heroku application management and deployment
* App Store Connect management (builds, TestFlight, reviews, IAPs)
* Google Play Store review management and analytics
* Google Workspace integration (Gmail, Calendar, Drive, Docs, Sheets, Slides, Forms, Tasks, Contacts, Chat) via optional [`workspace-mcp`](#google-workspace-mcp-optional) setup

## Installation

### Quick Start

The plugin bundles MCP servers that need `node` (for `npx`) and `uv` (for `uvx`). Most skills also use a CLI as a
fallback when the MCP isn't available. Install the **core** for everyone, then only the optional CLIs for the skills you
actually use.

#### macOS (Homebrew)

```bash
# Core (required)
brew install node uv gh jq
```

In Claude Code:

```text
claude plugin marketplace add cloud-officer/claude-code-plugin-dev
claude plugin install co-dev@cloud-officer
```

After install:

```bash
# Browser automation (Playwright MCP) — first run only
npx playwright install chromium
```

Optional service CLIs — install only what you need (each maps to a skill in the [Skills](#skills) table below):

```bash
brew install awscli google-cloud-sdk heroku newrelic-cli # cloud
brew tap ankitpokhrel/jira-cli && brew install jira-cli # jira
brew install mint && mint install zelentsov-dev/asc-mcp # iOS App Store Connect
```

Language servers (LSPs) — only what you write. The plugin declares LSPs for many languages, but each binary needs to be on your PATH for that language to activate:

```bash
# JavaScript/TypeScript (covers .ts/.tsx/.js/.jsx/.mts/.cts/.mjs/.cjs)
npm i -g typescript typescript-language-server

# Python (covers .py/.pyi)
brew install pyright

# Ruby (covers .rb/.rake/.gemspec/.ru)
gem install ruby-lsp

# Go (covers .go)
go install golang.org/x/tools/gopls@latest

# Bash/shell (covers .sh/.bash/.zsh)
npm i -g bash-language-server

# Rust (covers .rs)
rustup component add rust-analyzer

# YAML (covers .yml/.yaml; provides GitHub Actions/k8s schema validation)
npm i -g yaml-language-server

# Swift (covers .swift) — bundled with Xcode, no install needed if Xcode is installed

# PHP (covers .php)
npm i -g intelephense

# Kotlin (covers .kt/.kts) — JetBrains' official LSP, requires Java 17+
brew install JetBrains/utils/kotlin-lsp

# Java (covers .java)
brew install jdtls

# C/C++/Objective-C (covers .c/.cpp/.h/.hpp/.m/.mm) — bundled with llvm
brew install llvm

# Perl (covers .pl/.pm/.t)
npm i -g perlnavigator-server
```

Each LSP is **inert until matching files exist** — you only pay startup cost in repos that actually have those file
types.

#### Linux (Debian/Ubuntu)

```bash
# Core (required)
sudo apt update && sudo apt install -y curl jq git nodejs npm
curl -LsSf https://astral.sh/uv/install.sh | sh
# gh: https://github.com/cli/cli/blob/trunk/docs/install_linux.md
```

In Claude Code:

```text
claude plugin marketplace add cloud-officer/claude-code-plugin-dev
claude plugin install co-dev@cloud-officer
```

After install:

```bash
npx playwright install chromium
```

Optional service CLIs: each tool ships its own Linux installer — see official docs for `aws`, `gcloud`, `heroku`,
`newrelic`, `jira-cli`. App Store Connect is macOS-only.

Language servers (LSPs): the `npm`, `gem`, `go install`, and `rustup`
install commands listed under the macOS section are cross-platform. Use your distro's package manager for
`clangd` (often `clang-tools` or `llvm`) and `pyright` (or
`pip install pyright`). Swift LSP requires Xcode (macOS only).

#### Windows (Scoop)

```powershell
# Core (required)
scoop install nodejs uv gh jq
```

In Claude Code:

```text
claude plugin marketplace add cloud-officer/claude-code-plugin-dev
claude plugin install co-dev@cloud-officer
```

After install:

```powershell
npx playwright install chromium
```

Optional service CLIs: install via `scoop`, `winget`, or each tool's installer. App Store Connect is macOS-only.

Language servers (LSPs): the `npm`, `gem`, `go install`, and `rustup`
install commands listed under the macOS section work on Windows under their respective toolchains. Use
`scoop install llvm` for `clangd`. Swift LSP is macOS-only (requires Xcode).

### Configure Remote MCPs (optional)

Some skills use remote MCP servers (no local install). Add only the ones you'll use — `claude mcp add` defaults to
**local scope**, so each is registered for the current folder only rather than always-loaded for every project:

```bash
claude mcp add atlassian --transport http https://mcp.atlassian.com/v1/mcp
claude mcp add bigquery --transport http https://bigquery.googleapis.com/mcp
claude mcp add figma --transport http https://mcp.figma.com/mcp
claude mcp add monday --transport http https://mcp.monday.com/mcp --header 'Authorization: Bearer ${MONDAY_TOKEN}'
claude mcp add newrelic --transport http https://mcp.newrelic.com/mcp/
claude mcp add paypal --transport http https://mcp.paypal.com/http
```

Most prompt for OAuth on first use. **monday** is the exception — it authenticates with a personal access token instead
of OAuth, so it takes a Bearer header (the token stays a runtime `${MONDAY_TOKEN}` reference, never written into the
stored config). Skills that need a remote MCP fall back to a CLI when one exists — see the **Setup** column in the
[Skills](#skills) table for which skill needs which.

**Naming caveat for multi-tenant setups.** Each server name owns exactly one OAuth grant — Claude Code stores the access
token keyed by the name and reuses it for every call. If you work across multiple tenants of the same service (e.g. two
Atlassian sites, two New Relic accounts) from different folders on one machine, **give each instance a distinct name**
rather than reusing the bare name. Otherwise the OAuth token from whichever folder authorized first leaks into the other
folder, even though local-scope registration looks isolated.

```bash
# in ~/work/companya
claude mcp add atlassian-companya --transport http https://mcp.atlassian.com/v1/mcp

# in ~/work/companyb
claude mcp add atlassian-companyb --transport http https://mcp.atlassian.com/v1/mcp
```

This applies to every OAuth-at-connect remote MCP listed above (`atlassian`, `bigquery`, `figma`, `newrelic`, `paypal`) — not to stdio servers like
`google-workspace`, which select identity per-call via a parameter. If you rename instances, update the [Recommended Permissions](#recommended-permissions) entries to allow each variant.

### Google Workspace MCP (optional)

The
`co-dev` plugin focuses on engineering workflows (issues, PRs, code review, deployments). Pair it with the open-source [`workspace-mcp`](https://github.com/taylorwilsdon/google_workspace_mcp) server to add Gmail, Calendar, Drive, Docs, Sheets, Slides, Forms, Tasks, Contacts, and Google Chat — useful alongside several of this plugin's skills (e.g.
`weekly-dev-report` can email the report directly once Gmail is wired up, and
`sprint-summary` output can be pasted straight into a shared Doc or Calendar event).

This MCP is **not bundled** with the plugin — the steps below are independent and only needed if you want Workspace
access from Claude Code.

#### Step 1 — Google Cloud admin setup (one-time, per Workspace org)

Sign in at [console.cloud.google.com](https://console.cloud.google.com) **with your Workspace admin account** — the org
must be associated with the project, otherwise the **Internal** audience option will not appear later.

1. **Create the project** — top-bar dropdown → **New Project** → name `workspace-mcp`, **Organization** = your Workspace org (e.g. `yourcompany.com`). If your org doesn't appear, the Workspace and Cloud accounts aren't linked yet — fix at [admin.google.com](https://admin.google.com) → Account → Google Cloud.
2. **Enable the APIs** — APIs & Services → Library → enable each, one by one: Gmail, Google Calendar, Google Drive, Google Docs, Google Sheets, Google Slides, Google Forms, Tasks, People (for Contacts), Google Chat.
3. **Configure the OAuth consent screen** — APIs & Services → OAuth consent screen → **Get started**.
   * **App name:** `Workspace MCP`
   * **User support email:** your admin email
   * **Audience:** **Internal** (restricts auth to users in your org, no Google verification, no 100-user testing cap). If only **External** is offered, you signed in with the wrong account or created the project under No Organization.
   * **Developer contact:** your admin email
4. **Create the OAuth client credentials** — APIs & Services → Credentials → **Create Credentials → OAuth Client ID**.
   * **Application type:** **Desktop app** (Web application will not work — it causes redirect URI errors during auth)
   * **Name:** `workspace-mcp-desktop`
5. Copy the **Client ID** (`...apps.googleusercontent.com`) and **Client Secret** (`GOCSPX-...`). Treat the secret like a password; never commit it.

The same Client ID and Secret work for every user in the org since the consent screen is Internal. **Multiple Workspace
organizations** (e.g. a personal org and a separate company org) each need their own Cloud project and credentials — the
same Client ID cannot be reused across orgs.

#### Step 2 — Per-machine install

**macOS:**

```bash
brew install uv direnv
```

Hook direnv into your shell — add to `~/.zshrc` (once, then `source ~/.zshrc`):

```bash
eval "$(direnv hook zsh)"
```

Then in the directory where you'll launch Claude Code, create an `.envrc` file with everything workspace-mcp needs:

```bash
export GOOGLE_OAUTH_CLIENT_ID="your-client-id"
export GOOGLE_OAUTH_CLIENT_SECRET="your-client-secret"
export USER_GOOGLE_EMAIL="you@yourdomain.com"
export PORT="8765" # any unused port — see "Multi-account & multi-session" below for why this matters
```

Approve it with `direnv allow` (run from that directory). direnv loads the vars whenever you
`cd` in and unloads them when you `cd` out — keep one
`.envrc` per workspace org / per concurrently-running session, each with its own `USER_GOOGLE_EMAIL` and
`PORT`. See [Multi-Account Setups (direnv)](#multi-account-setups-direnv) below for the caveats around switching mid-session.

**Windows (PowerShell, no admin needed):**

```powershell
powershell -ExecutionPolicy ByPass -c "irm https://astral.sh/uv/install.ps1 | iex"
[System.Environment]::SetEnvironmentVariable("GOOGLE_OAUTH_CLIENT_ID", "your-client-id", "User")
[System.Environment]::SetEnvironmentVariable("GOOGLE_OAUTH_CLIENT_SECRET", "your-client-secret", "User")
[System.Environment]::SetEnvironmentVariable("USER_GOOGLE_EMAIL", "you@yourdomain.com", "User")
[System.Environment]::SetEnvironmentVariable("PORT", "8765", "User")
```

Close and reopen PowerShell so the new env vars take effect. Git for Windows must also be installed (Claude Code uses Git Bash internally) — get it from [git-scm.com](https://git-scm.com/download/win) with default options.

User-scope env vars are global on Windows, so if you need a second workspace account or run multiple Claude sessions concurrently, override
`USER_GOOGLE_EMAIL` and `PORT` per session before launching Claude (
`$env:USER_GOOGLE_EMAIL="..."; $env:PORT="8766"; claude`).

#### Step 3 — Add the MCP server

Run this **from the directory** where you want the server active (the default scope is local — the registration is tied
to that working directory, not your whole user account). The server inherits
`GOOGLE_OAUTH_CLIENT_ID`, `GOOGLE_OAUTH_CLIENT_SECRET`, `USER_GOOGLE_EMAIL`, and
`PORT` from the shell environment Claude Code is launched in, so no `-e` flags are needed:

```bash
claude mcp add google-workspace -- uvx workspace-mcp --tool-tier complete
```

(Add `--scope user` if you want the server available globally in every directory; add
`--scope project` to write it to a shared `.mcp.json` in the repo — but never with hard-coded credentials, since
`.mcp.json` typically gets committed.)

On first use, a browser window opens for OAuth — sign in with the matching Google account and approve the scopes. Tokens cache locally:

* macOS: `~/.google_workspace_mcp/credentials/`
* Windows: `%USERPROFILE%\.google_workspace_mcp\credentials\`

No browser prompt on subsequent startups. To **update** workspace-mcp: `uv cache clean` — Claude Code fetches the latest
version on next startup. To **revoke** access: delete the credentials directory above, or remove the `Workspace MCP` app
at [myaccount.google.com/permissions](https://myaccount.google.com/permissions). To **rotate** the Client Secret if it
leaks: APIs & Services → Credentials → `workspace-mcp-desktop` → **Reset Secret**, then update env vars and re-auth.

#### Multi-account & multi-session port assignment (important)

The workspace-mcp server binds a local port for its OAuth callback (default `8000`). **Every workspace-mcp instance
running concurrently on the machine needs a unique port.** There are two scenarios where this matters — most people
anticipate the first but not the second:

1. **Multiple accounts on the same machine** (e.g. personal + company in different folders). Each folder's
   `.envrc` (macOS/direnv) — or each session's PowerShell overrides (Windows) — needs its own
   `PORT`, otherwise both instances try to bind `8000` and the second one fails.
2. **Multiple Claude Code sessions running at the same time, even from different directories.** Per-directory env vars
   don't help if two sessions still resolve to the same `PORT` — every running Claude Code session spawns its own MCP
   processes, and if any two of them point at the same port, the later one's workspace-mcp won't start. Even with a
   single account per directory, every directory whose Claude session might run *concurrently* with another needs its
   own port.

Pick ports above 8000 (e.g. `8765`, `8766`, `8767`, …) and assign one per account
*and* per concurrently-running session. The setup is then **one `.envrc` per directory** plus a single
`claude mcp add` from each. Example with two accounts on macOS:

```bash
# ~/work/personal-stuff/.envrc
export GOOGLE_OAUTH_CLIENT_ID="personal-client-id"
export GOOGLE_OAUTH_CLIENT_SECRET="personal-secret"
export USER_GOOGLE_EMAIL="you@personal.com"
export PORT="8765"
```

```bash
# ~/work/cloudofficer/.envrc
export GOOGLE_OAUTH_CLIENT_ID="company-client-id"
export GOOGLE_OAUTH_CLIENT_SECRET="company-secret"
export USER_GOOGLE_EMAIL="you@company.com"
export PORT="8766"
```

Run `direnv allow` once in each directory, then register the server from each (no
`-e` flags — credentials and port come from `.envrc`):

```bash
cd ~/work/personal-stuff && claude mcp add google-workspace -- uvx workspace-mcp --tool-tier complete
cd ~/work/cloudofficer && claude mcp add google-workspace -- uvx workspace-mcp --tool-tier complete
```

The port only matters during the OAuth callback handshake; it does not need to be reachable from outside the machine. Verify everything is registered (run from inside each directory):

```bash
claude mcp list
```

Sample prompts once it's up:

```text
List my unread emails
Search drive for the Q3 proposal
Create a calendar event for tomorrow at 2pm
```

### Complementary Plugins (optional)

These official plugins from the `claude-plugins-official` marketplace pair well with
`co-dev`. They're independent — install only the ones relevant to your work. The
`claude-plugins-official` marketplace is auto-installed by Claude Code, so no `marketplace add` step is needed.

| Plugin                 | What it adds                                                                                                                                                                          | Install                                                                                                               |
|------------------------|---------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------|-----------------------------------------------------------------------------------------------------------------------|
| `plugin-dev`           | Toolkit for developing Claude Code plugins — 7 skills (hooks, MCP, structure, settings, commands, agents, skills) + validators                                                        | `claude plugin install plugin-dev@claude-plugins-official`                                                            |
| `claude-md-management` | Audit and revise CLAUDE.md files based on session learnings                                                                                                                           | `claude plugin install claude-md-management@claude-plugins-official`                                                  |
| `claude-code-setup`    | Analyze a codebase and recommend Claude Code automations to add                                                                                                                       | `claude plugin install claude-code-setup@claude-plugins-official`                                                     |
| `skill-creator`        | Guided skill creation workflow (overlaps with `plugin-dev`'s skill-development skill — pick one)                                                                                      | `claude plugin install skill-creator@claude-plugins-official`                                                         |
| `code-simplifier`      | Refactor recently-modified code for clarity, DRY, and consistency                                                                                                                     | `claude plugin install code-simplifier@claude-plugins-official`                                                       |
| `code-review`          | Lean automated PR review with confidence scoring (alternative to `co-dev`'s `/code-review-deep`, which is broader and not PR-bound)                                                   | `claude plugin install code-review@claude-plugins-official`                                                           |
| `security-guidance`    | Proactive security hints via PreToolUse hooks (complements `code-review-deep`'s reactive audit)                                                                                       | `claude plugin install security-guidance@claude-plugins-official`                                                     |
| `frontend-design`      | Opinionated patterns for production-grade frontend interfaces                                                                                                                         | `claude plugin install frontend-design@claude-plugins-official`                                                       |
| `superpowers`          | Brainstorming and subagent-driven workflows                                                                                                                                           | `claude plugin install superpowers@claude-plugins-official`                                                           |
| `agent-sdk-dev`        | Toolkit for building apps on the Anthropic **Agent SDK** (a different stack from Claude Code plugins)                                                                                 | `claude plugin install agent-sdk-dev@claude-plugins-official`                                                         |
| `document-skills`      | Anthropic's `xlsx`/`docx`/`pptx`/`pdf` skills — deliver `co-dev` reports (`weekly-dev-report`, `sprint-summary`, `query-db`, …) as Office/PDF files. Separate marketplace (see note). | `claude plugin marketplace add anthropics/skills` then `claude plugin install document-skills@anthropic-agent-skills` |
| `stripe`               | Stripe's official plugin — integration skills (best practices, SDK/API upgrades) **plus** Stripe's own MCP + agent tools                                                              | `claude plugin install stripe@claude-plugins-official`                                                                |
| `vercel`               | Vercel's official plugin — deployment ops via the Vercel MCP, plus `/deploy` `/env` `/status` commands and a deployment-expert agent                                                  | `claude plugin install vercel@claude-plugins-official`                                                                |
| `huggingface-skills`   | Hugging Face's official skills — build, train, evaluate, and use open-source models, datasets, and Spaces (HF CLI, trainers, evaluation, Gradio). For ML/model work.                  | `claude plugin install huggingface-skills@claude-plugins-official`                                                    |
| `sentry`               | Sentry error monitoring — access error reports, analyze stack traces, search issues, debug production errors, and set up Sentry SDKs                                                  | `claude plugin install sentry@claude-plugins-official`                                                                |
| `azure`                | Microsoft's official Azure plugin — bundles the Azure MCP + ~28 ops skills: list resources, deploy, validate, diagnose, RBAC/Entra, and optimize costs across 50+ services            | `claude plugin install azure@claude-plugins-official`                                                                 |
| `firebase`             | Google Firebase plugin — bundles the Firebase MCP to manage Firestore, auth, cloud functions, hosting, and storage. Complements `co-dev`'s `crashlytics` skill                        | `claude plugin install firebase@claude-plugins-official`                                                              |

> The `stripe` plugin bundles Stripe's own MCP and agent tools, so you do **not** need a separate
> `claude mcp add stripe`. For local webhook
> testing you can optionally add the Stripe CLI (used by Stripe's plugin and your app dev, not by `co-dev`
> itself): `brew install stripe/stripe-cli/stripe` then `stripe login` (`stripe listen`, `stripe trigger`).
>
> The `vercel` plugin bundles the Vercel MCP, so no separate `claude mcp add vercel` is needed. That MCP is
> read-only (docs, list projects/deployments, logs); write operations (`deploy`, `promote`, env changes) run
> through the `vercel` CLI — install it too: `brew install vercel-cli` then `vercel login`.
>
> The `huggingface-skills` plugin's Hub operations use the Hugging Face CLI — install it with
> `brew install hf` then `hf auth login`. Its training and evaluation skills additionally need a Python ML
> environment (torch, transformers, TRL) specific to your project.
>
> The `sentry` plugin is skill-based (SDK setup + debug workflows). For CLI operations — release tracking and
> source-map uploads — install the Sentry CLI: `brew install sentry-cli` (also packaged as the `sentry-cli`
> plugin).
>
> The `azure` plugin bundles the Azure MCP (`@azure/mcp`), so no separate `claude mcp add` is needed. It
> authenticates with your Azure CLI credentials — install it with `brew install azure-cli` then `az login`.
>
> The `firebase` plugin bundles the Firebase MCP, so no separate `claude mcp add` is needed. Its CLI
> operations (deploy, emulators, security rules) use the Firebase CLI — install it with
> `brew install firebase-cli` then `firebase login`. It complements `co-dev`'s `crashlytics` skill (crash
> data via BigQuery) rather than replacing it.
>
> `document-skills` lives on Anthropic's own `anthropic-agent-skills` marketplace (not
> `claude-plugins-official`), so it needs the one-time `claude plugin marketplace add` step above. It is
> **not** open source — review each skill's `LICENSE.txt`. Optional: Claude Code can already generate
> these formats on demand via open-source libraries (`openpyxl`, `python-docx`, `python-pptx`,
> `reportlab`); the skills mainly add standardized, repeatable formatting.

### Configure Environment Variables

All credentials are read from **environment variables** — they are never stored in the plugin. Each skill's required
vars are listed in the **Setup** column of the [Skills](#skills) table below.

For switching between accounts/projects per directory, see the [Multi-Account Setups](#multi-account-setups-direnv) section.

### Multi-Account Setups (direnv)

Every account-bound MCP server in this plugin reads its credentials from environment variables — no credentials are stored in
`.mcp.json`. This makes [direnv](https://direnv.net/) a natural fit for switching between accounts (different AWS profiles, separate Postgres instances, multiple GitHub orgs, etc.) by setting per-directory env vars in an
`.envrc` file.

**Important caveat:** Claude Code spawns MCP servers **once, at startup**, and they inherit the shell environment at
that moment. The implication:

* ✓ Works: `cd ~/project-a && claude` — MCPs pick up project-a's env vars from `.envrc`. Quit,
  `cd ~/project-b && claude` — MCPs pick up project-b's env vars.
* ✗ Does not work: starting Claude Code in one project, then
  `cd`-ing to another mid-session. The already-running MCP servers keep the original env vars and continue talking to the original account.

**Bottom line:** to switch accounts, quit Claude Code and relaunch it from the target directory. direnv handles the
rest.

### Recommended Permissions

This plugin bundles several MCP servers. By default, Claude Code will prompt for permission each time an MCP tool is called. To auto-approve these tools, add the following entries to the
`permissions.allow` array in your `~/.claude/settings.json`:

```json
{
  "permissions": {
    "allow": [
      "mcp__appstore__*",
      "mcp__aws__*",
      "mcp__chrome-devtools__*",
      "mcp__context7__*",
      "mcp__fetch__*",
      "mcp__gcloud__*",
      "mcp__github__*",
      "mcp__mongodb__*",
      "mcp__mysql__*",
      "mcp__playstore__*",
      "mcp__playwright__*",
      "mcp__postgres__*",
      "mcp__redis__*"
    ]
  }
}
```

If you also use remote MCP servers (see [Configure Remote MCPs](#configure-remote-mcps-optional)), add their patterns too:

```json
{
  "permissions": {
    "allow": [
      "mcp__atlassian__*",
      "mcp__bigquery__*",
      "mcp__figma__*",
      "mcp__monday__*",
      "mcp__newrelic__*",
      "mcp__paypal__*"
    ]
  }
}
```

**Note:** These entries merge with your existing
`allow` list — you don't need to replace it. Only add entries for the MCP servers you actually use.

**Renamed instances** (see [Naming caveat for multi-tenant setups](#configure-remote-mcps-optional)) need their own
permission entry — e.g. `mcp__atlassian-companya__*` and `mcp__atlassian-companyb__*` instead of (or in addition to) the
bare `mcp__atlassian__*`.

## Usage

### Skills

These skills are automatically available to Claude. The **Setup** column lists what you need to configure for each — env
vars, one-time auth commands, or remote MCP additions. Skills with no setup work out of the box.

| Skill                    | Description                                                                                                             | Setup                                                                                                                                                                                                                                               |
|--------------------------|-------------------------------------------------------------------------------------------------------------------------|-----------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------|
| `analyze-db`             | Generate docs/db.md with database schema docs                                                                           | DB-specific env vars — see `query-db` row                                                                                                                                                                                                           |
| `appstore`               | Manage App Store Connect (builds, TestFlight, reviews, IAPs)                                                            | macOS-only. `mint install zelentsov-dev/asc-mcp`. Env: `ASC_KEY_ID`, `ASC_ISSUER_ID`, `ASC_PRIVATE_KEY_PATH` (path to `.p8`). [Create API key](https://appstoreconnect.apple.com/access/integrations/api)                                           |
| `aws`                    | Manage AWS infrastructure and services                                                                                  | `aws configure`, or env: `AWS_PROFILE` *(or)* `AWS_ACCESS_KEY_ID` + `AWS_SECRET_ACCESS_KEY` + `AWS_REGION`                                                                                                                                          |
| `code-review-deep`       | Exhaustive multi-phase code audit using parallel agents                                                                 | None — uses bundled `github` + `context7` MCPs                                                                                                                                                                                                      |
| `create-issue`           | Create GitHub or Jira issues with proper templates                                                                      | GitHub: `GITHUB_PERSONAL_ACCESS_TOKEN` *(or)* `gh auth login`. Jira: atlassian remote MCP *(or)* `jira init`                                                                                                                                        |
| `create-pr`              | Generate PR content, open the PR, return to default branch                                                              | GitHub: `gh auth login` *(or)* `GITHUB_PERSONAL_ACCESS_TOKEN`                                                                                                                                                                                       |
| `crashlytics`            | Query Firebase Crashlytics crash data from BigQuery                                                                     | bigquery remote MCP *(or)* `gcloud auth application-default login`. Env: `BQ_PROJECT`, `BQ_CRASHLYTICS_DATASET`                                                                                                                                     |
| `doc-tracker-coverage`   | Verify a Google Doc section's items are tracked in Monday/Jira                                                          | Google Doc read: Workspace Docs MCP *(or)* `claude.ai Google Drive` MCP *(or)* published-to-web doc. monday: `claude mcp add monday` (`${MONDAY_TOKEN}`). Jira: atlassian remote MCP *(or)* `jira init` Optional env: `DOC_TRACKER_COVERAGE_TEAMS`  |
| `gcloud`                 | Manage Google Cloud infrastructure and services                                                                         | `gcloud auth login && gcloud config set project <id>`. Optional: `CLOUDSDK_ACTIVE_CONFIG_NAME` for named configs                                                                                                                                    |
| `heroku`                 | Manage Heroku apps, dynos, logs, and databases                                                                          | `heroku login`                                                                                                                                                                                                                                      |
| `loco`                   | Manage Loco translation assets (create, delete, scan)                                                                   | Env: `LOCO_API_KEY` *(or)* per-project `LOCO_API_KEY_<PROJECT>` (e.g. `LOCO_API_KEY_IOS`)                                                                                                                                                           |
| `migrate-code`           | Port a codebase to another language/framework via a six-step rulebook/translate/compile/test/verify engine              | None — uses the repo's own build/test commands + bundled `context7` MCP                                                                                                                                                                             |
| `monday`                 | Manage monday.com boards, items, groups, columns, and updates                                                           | `claude mcp add monday` with a `${MONDAY_TOKEN}` Bearer header — see [Configure Remote MCPs](#configure-remote-mcps-optional). Token from avatar → Developers → My access tokens. MCP-only, no CLI fallback                                         |
| `monday-weekly-report`   | Weekly project-status report from monday.com board(s)                                                                   | `claude mcp add monday` with a `${MONDAY_TOKEN}` Bearer header (see `monday` above). For attribution enable dynamic API. Optional env: `MONDAY_WEEKLY_REPORT_BOARDS`, `MONDAY_WEEKLY_REPORT_TO` (required for `--send`), `MONDAY_WEEKLY_REPORT_CC`  |
| `newrelic`               | Query New Relic observability data, alerts, and logs                                                                    | newrelic remote MCP *(or)* `newrelic profile add --name default --apiKey NRAK-... --accountId ...`                                                                                                                                                  |
| `paypal`                 | Manage PayPal invoices, payments, and disputes                                                                          | paypal remote MCP **(required — no CLI fallback)**                                                                                                                                                                                                  |
| `playstore`              | Fetch, analyze, and respond to Google Play reviews                                                                      | Env: `GOOGLE_APPLICATION_CREDENTIALS` (path to service-account JSON; needs Google Play Developer API + Play Console access). Requires `uv`/`uvx` on PATH.                                                                                           |
| `query-db`               | Query databases using natural language via CLI                                                                          | Per DB: PG (`PGHOST`, `PGPORT`, `PGUSER`, `PGPASSWORD`, `PGDATABASE`) · MySQL (`MYSQL_*`) · Mongo (`MDB_MCP_CONNECTION_STRING`) · Redis (`REDIS_URL`) · SQLite (`SQLITE_DB`) · BigQuery (`BQ_PROJECT`, `BQ_DATASETS`) · ES (`ES_URL`, `ES_API_KEY`) |
| `review-architecture`    | Review or create docs/architecture.md                                                                                   | None — uses bundled `fetch` + `context7` MCPs                                                                                                                                                                                                       |
| `review-copy`            | Audit user-facing copy — microcopy, form/help text, voice & tone, i18n readiness, marketing; writes docs/copy-review.md | None — uses `git` + optional `loco` handoff for i18n keys                                                                                                                                                                                           |
| `review-design`          | Compare UI code against Figma designs (Android, iOS, web)                                                               | figma remote MCP **(required — no CLI fallback)**                                                                                                                                                                                                   |
| `review-ownership-map`   | Code ownership & bus factor from git history — flags single points of failure on sensitive code                         | None — read-only git analysis                                                                                                                                                                                                                       |
| `review-readme`          | Review or create README.md to match standards                                                                           | None — uses bundled `fetch` + `context7` MCPs                                                                                                                                                                                                       |
| `review-seo`             | Audit SEO + GEO + front-end web quality (perf, Core Web Vitals, accessibility) for a codebase and/or live URL           | None — uses `curl` + bundled `fetch` + `chrome-devtools` MCPs                                                                                                                                                                                       |
| `review-threat-model`    | Repo-specific STRIDE threat model — trust boundaries, abuse paths, ISO 27001 evidence                                   | None — read-only code analysis                                                                                                                                                                                                                      |
| `review-user-guide`      | Review or create docs/user-guide.md with user documentation                                                             | None — uses bundled `fetch` + `context7` MCPs                                                                                                                                                                                                       |
| `run-linters`            | Run linters and fix any issues found                                                                                    | None — runs project linters already configured in the repo                                                                                                                                                                                          |
| `sprint-summary`         | Summarize sprint items grouped by repo in ~3-day blocks                                                                 | atlassian remote MCP *(or)* `jira init`                                                                                                                                                                                                             |
| `verify-resolved-issues` | Audit resolved/fixed/done issues, verify fixes, close or reopen                                                         | GitHub: `gh auth login` *(or)* `GITHUB_PERSONAL_ACCESS_TOKEN`. Jira: atlassian remote MCP *(or)* `jira init`                                                                                                                                        |
| `weekly-dev-report`      | Weekly dev activity report from Jira sprint + GitHub                                                                    | atlassian remote MCP *(or)* `jira init`. GitHub: `gh auth login`. Optional env: `WEEKLY_DEV_REPORT_TO` (required for `--send`), `WEEKLY_DEV_REPORT_CC`, `GITHUB_USERNAME_MAP`                                                                       |
| `work-issue`             | Implement GitHub/Jira issue(s) end-to-end and open a PR                                                                 | GitHub: `gh auth login` *(or)* `GITHUB_PERSONAL_ACCESS_TOKEN`. Jira: atlassian remote MCP *(or)* `jira init`                                                                                                                                        |
| `write-tests`            | Generate and run tests in the repo's own framework (unit/integration/e2e); enforces 80/80 line/branch coverage          | None — uses the test runner already in the repo (RSpec, pytest, Jest/Vitest, go test, dotnet, xcodebuild, gradle, PHPUnit/Pest, cargo, ctest, …)                                                                                                    |

### Language Servers (LSPs)

The plugin declares LSPs for 13 languages. Each is **idle until you open a matching file** in your workspace — opening
`app.ts` activates the TypeScript LSP, opening `lib.go` activates
`gopls`, etc. No LSP runs in a repo that doesn't have matching file extensions.

**Why they matter:** an active LSP gives Claude **real type information** from your installed dependencies — function
signatures, references, type definitions, completions — instead of inferring from source code. This is what stops Claude
from hallucinating method names on third-party libraries or guessing the shape of a return type. Any skill that touches
code (`code-review-deep`, `migrate-code`, `run-linters`, `work-issue`, `write-tests`, `create-pr`, the deep-analysis
agents) becomes meaningfully more accurate.

**Activation rules:**

* LSPs spawn on demand when matching file extensions exist in the workspace, and stop when Claude Code exits.
* If the LSP binary is not on your
  `PATH`, that language silently falls back to source-only analysis — the rest of the plugin is unaffected.
* Each LSP communicates over stdio with Claude Code's LSP host; you don't interact with them directly.
* They are **independent of the bundled MCP servers** — different protocol, different lifecycle, different purpose. MCPs
  give Claude tools (e.g., "search GitHub", "query a database"). LSPs give Claude knowledge of your code.

**Bundled LSPs:**

| Language               | LSP binary                           | File extensions                                                       |
|------------------------|--------------------------------------|-----------------------------------------------------------------------|
| TypeScript /JavaScript | `typescript-language-server`         | `.ts`, `.tsx`, `.js`, `.jsx`, `.mts`, `.cts`, `.mjs`, `.cjs`          |
| Python                 | `pyright-langserver`                 | `.py`, `.pyi`                                                         |
| Ruby                   | `ruby-lsp`                           | `.rb`, `.rake`, `.gemspec`, `.ru`                                     |
| Go                     | `gopls`                              | `.go`                                                                 |
| Bash                   | `bash-language-server`               | `.sh`, `.bash`, `.zsh`                                                |
| Rust                   | `rust-analyzer`                      | `.rs`                                                                 |
| YAML                   | `yaml-language-server`               | `.yml`, `.yaml` (provides GitHub Actions / k8s schema validation)     |
| Swift                  | `sourcekit-lsp` (bundled with Xcode) | `.swift`                                                              |
| PHP                    | `intelephense`                       | `.php`                                                                |
| Kotlin                 | `kotlin-lsp` (JetBrains, pre-alpha)  | `.kt`, `.kts` — JVM-only Gradle/Maven projects; KMP not yet supported |
| Java                   | `jdtls`                              | `.java`                                                               |
| C / C++ / Objective-C  | `clangd`                             | `.c`, `.cpp`, `.cc`, `.cxx`, `.h`, `.hpp`, `.hh`, `.hxx`, `.m`, `.mm` |
| Perl                   | `perlnavigator`                      | `.pl`, `.pm`, `.t`                                                    |

Install commands per LSP are in the [macOS Quick Start](#macos-homebrew) section. **Install only the LSPs for languages
you actually write** — each one adds a small startup cost only when its file types appear in a workspace.

### Local Development

```bash
claude --plugin-dir /path/to/claude-code-plugin-dev
```

## Contributing

We love your input! We want to make contributing to this project as easy and transparent as possible, whether it's:

* Reporting a bug
* Discussing the current state of the code
* Submitting a fix
* Proposing new features
* Becoming a maintainer

Pull requests are the best way to propose changes to the codebase. We actively welcome your pull requests:

1. Fork the repo and create your branch from `master`.
2. If you've added code that should be tested, add tests. Ensure the test suite passes.
3. Update the documentation.
4. Make sure your code lints.
5. Issue that pull request!

When you submit code changes, your submissions are understood to be under the same [License](https://github.com/Cloud-Officer/claude-code-plugin-dev/tree/HEAD/LICENSE) that covers the project. Feel free to contact the maintainers if that's a concern.

claude-code-plugin-dev

Claude Code plugin for development workflow automation.

Introduction
Installation
Usage
Contributing

Introduction

This plugin provides development workflow automation for Claude Code.

Features

Issue tracking integration (GitHub Issues and Jira)
Automated deep code reviews using parallel agents
PR generation with commit messages
Linting and code quality checks
Automated test generation with coverage enforcement across 10 languages (RSpec, PHPUnit/Pest, pytest, Jest/Vitest, xUnit, Swift Testing/XCUITest, Kotest, Go, GoogleTest, Rust)
Codebase migration between languages and frameworks via a six-step rulebook/translate/compile/test/verify engine
Security threat modeling (STRIDE) and code-ownership / bus-factor analysis, ISO 27001-aligned
Database schema documentation and natural language querying
Project documentation review (README, architecture, user guide)
User-facing copy audits (microcopy, voice & tone, i18n readiness)
SEO / GEO and front-end web quality audits (performance, Core Web Vitals, accessibility)
Sprint work summaries grouped by repository
Weekly development and project-status reports from Jira, GitHub, and monday.com
Tracker coverage checks — verify a planning doc's items exist in monday.com and/or Jira
Translation asset management via Loco (localise.biz) API
monday.com board, item, and work-management operations
Firebase Crashlytics crash analysis via BigQuery
Figma design-to-code review for Android, iOS, and web
PayPal payment management
AWS infrastructure management and documentation search
Google Cloud infrastructure management via gcloud
New Relic observability data and alert management
Heroku application management and deployment
App Store Connect management (builds, TestFlight, reviews, IAPs)
Google Play Store review management and analytics
Google Workspace integration (Gmail, Calendar, Drive, Docs, Sheets, Slides, Forms, Tasks, Contacts, Chat) via optional workspace-mcp setup

Installation

Quick Start

The plugin bundles MCP servers that need node (for npx) and uv (for uvx). Most skills also use a CLI as a fallback when the MCP isn't available. Install the core for everyone, then only the optional CLIs for the skills you actually use.

macOS (Homebrew)

# Core (required)
brew install node uv gh jq

In Claude Code:

claude plugin marketplace add cloud-officer/claude-code-plugin-dev
claude plugin install co-dev@cloud-officer

After install:

# Browser automation (Playwright MCP) — first run only
npx playwright install chromium

Optional service CLIs — install only what you need (each maps to a skill in the Skills table below):

brew install awscli google-cloud-sdk heroku newrelic-cli # cloud
brew tap ankitpokhrel/jira-cli && brew install jira-cli # jira
brew install mint && mint install zelentsov-dev/asc-mcp # iOS App Store Connect

Language servers (LSPs) — only what you write. The plugin declares LSPs for many languages, but each binary needs to be on your PATH for that language to activate:

# JavaScript/TypeScript (covers .ts/.tsx/.js/.jsx/.mts/.cts/.mjs/.cjs)
npm i -g typescript typescript-language-server

# Python (covers .py/.pyi)
brew install pyright

# Ruby (covers .rb/.rake/.gemspec/.ru)
gem install ruby-lsp

# Go (covers .go)
go install golang.org/x/tools/gopls@latest

# Bash/shell (covers .sh/.bash/.zsh)
npm i -g bash-language-server

# Rust (covers .rs)
rustup component add rust-analyzer

# YAML (covers .yml/.yaml; provides GitHub Actions/k8s schema validation)
npm i -g yaml-language-server

# Swift (covers .swift) — bundled with Xcode, no install needed if Xcode is installed

# PHP (covers .php)
npm i -g intelephense

# Kotlin (covers .kt/.kts) — JetBrains' official LSP, requires Java 17+
brew install JetBrains/utils/kotlin-lsp

# Java (covers .java)
brew install jdtls

# C/C++/Objective-C (covers .c/.cpp/.h/.hpp/.m/.mm) — bundled with llvm
brew install llvm

# Perl (covers .pl/.pm/.t)
npm i -g perlnavigator-server

Each LSP is inert until matching files exist — you only pay startup cost in repos that actually have those file types.

Linux (Debian/Ubuntu)

# Core (required)
sudo apt update && sudo apt install -y curl jq git nodejs npm
curl -LsSf https://astral.sh/uv/install.sh | sh
# gh: https://github.com/cli/cli/blob/trunk/docs/install_linux.md

In Claude Code:

claude plugin marketplace add cloud-officer/claude-code-plugin-dev
claude plugin install co-dev@cloud-officer

After install:

npx playwright install chromium

Optional service CLIs: each tool ships its own Linux installer — see official docs for aws, gcloud, heroku, newrelic, jira-cli. App Store Connect is macOS-only.

Language servers (LSPs): the npm, gem, go install, and rustup install commands listed under the macOS section are cross-platform. Use your distro's package manager for clangd (often clang-tools or llvm) and pyright (or pip install pyright). Swift LSP requires Xcode (macOS only).

Windows (Scoop)

# Core (required)
scoop install nodejs uv gh jq

In Claude Code:

claude plugin marketplace add cloud-officer/claude-code-plugin-dev
claude plugin install co-dev@cloud-officer

After install:

npx playwright install chromium

Optional service CLIs: install via scoop, winget, or each tool's installer. App Store Connect is macOS-only.

Language servers (LSPs): the npm, gem, go install, and rustup install commands listed under the macOS section work on Windows under their respective toolchains. Use scoop install llvm for clangd. Swift LSP is macOS-only (requires Xcode).

Configure Remote MCPs (optional)

Some skills use remote MCP servers (no local install). Add only the ones you'll use — claude mcp add defaults to local scope, so each is registered for the current folder only rather than always-loaded for every project:

claude mcp add atlassian --transport http https://mcp.atlassian.com/v1/mcp
claude mcp add bigquery --transport http https://bigquery.googleapis.com/mcp
claude mcp add figma --transport http https://mcp.figma.com/mcp
claude mcp add monday --transport http https://mcp.monday.com/mcp --header 'Authorization: Bearer ${MONDAY_TOKEN}'
claude mcp add newrelic --transport http https://mcp.newrelic.com/mcp/
claude mcp add paypal --transport http https://mcp.paypal.com/http

Most prompt for OAuth on first use. monday is the exception — it authenticates with a personal access token instead of OAuth, so it takes a Bearer header (the token stays a runtime ${MONDAY_TOKEN} reference, never written into the stored config). Skills that need a remote MCP fall back to a CLI when one exists — see the Setup column in the Skills table for which skill needs which.

Naming caveat for multi-tenant setups. Each server name owns exactly one OAuth grant — Claude Code stores the access token keyed by the name and reuses it for every call. If you work across multiple tenants of the same service (e.g. two Atlassian sites, two New Relic accounts) from different folders on one machine, give each instance a distinct name rather than reusing the bare name. Otherwise the OAuth token from whichever folder authorized first leaks into the other folder, even though local-scope registration looks isolated.

# in ~/work/companya
claude mcp add atlassian-companya --transport http https://mcp.atlassian.com/v1/mcp

# in ~/work/companyb
claude mcp add atlassian-companyb --transport http https://mcp.atlassian.com/v1/mcp

This applies to every OAuth-at-connect remote MCP listed above (atlassian, bigquery, figma, newrelic, paypal) — not to stdio servers like google-workspace, which select identity per-call via a parameter. If you rename instances, update the Recommended Permissions entries to allow each variant.

Google Workspace MCP (optional)

The co-dev plugin focuses on engineering workflows (issues, PRs, code review, deployments). Pair it with the open-source workspace-mcp server to add Gmail, Calendar, Drive, Docs, Sheets, Slides, Forms, Tasks, Contacts, and Google Chat — useful alongside several of this plugin's skills (e.g. weekly-dev-report can email the report directly once Gmail is wired up, and sprint-summary output can be pasted straight into a shared Doc or Calendar event).

This MCP is not bundled with the plugin — the steps below are independent and only needed if you want Workspace access from Claude Code.

Step 1 — Google Cloud admin setup (one-time, per Workspace org)

Sign in at console.cloud.google.com with your Workspace admin account — the org must be associated with the project, otherwise the Internal audience option will not appear later.

Create the project — top-bar dropdown → New Project → name workspace-mcp, Organization = your Workspace org (e.g. yourcompany.com). If your org doesn't appear, the Workspace and Cloud accounts aren't linked yet — fix at admin.google.com → Account → Google Cloud.
Enable the APIs — APIs & Services → Library → enable each, one by one: Gmail, Google Calendar, Google Drive, Google Docs, Google Sheets, Google Slides, Google Forms, Tasks, People (for Contacts), Google Chat.
Configure the OAuth consent screen — APIs & Services → OAuth consent screen → Get started.
- App name: Workspace MCP
- User support email: your admin email
- Audience: Internal (restricts auth to users in your org, no Google verification, no 100-user testing cap). If only External is offered, you signed in with the wrong account or created the project under No Organization.
- Developer contact: your admin email
Create the OAuth client credentials — APIs & Services → Credentials → Create Credentials → OAuth Client ID.
- Application type: Desktop app (Web application will not work — it causes redirect URI errors during auth)
- Name: workspace-mcp-desktop
Copy the Client ID (...apps.googleusercontent.com) and Client Secret (GOCSPX-...). Treat the secret like a password; never commit it.

The same Client ID and Secret work for every user in the org since the consent screen is Internal. Multiple Workspace organizations (e.g. a personal org and a separate company org) each need their own Cloud project and credentials — the same Client ID cannot be reused across orgs.

Step 2 — Per-machine install

macOS:

brew install uv direnv

Hook direnv into your shell — add to ~/.zshrc (once, then source ~/.zshrc):

eval "$(direnv hook zsh)"

Then in the directory where you'll launch Claude Code, create an .envrc file with everything workspace-mcp needs:

export GOOGLE_OAUTH_CLIENT_ID="your-client-id"
export GOOGLE_OAUTH_CLIENT_SECRET="your-client-secret"
export USER_GOOGLE_EMAIL="you@yourdomain.com"
export PORT="8765" # any unused port — see "Multi-account & multi-session" below for why this matters

Approve it with direnv allow (run from that directory). direnv loads the vars whenever you cd in and unloads them when you cd out — keep one .envrc per workspace org / per concurrently-running session, each with its own USER_GOOGLE_EMAIL and PORT. See Multi-Account Setups (direnv) below for the caveats around switching mid-session.

Windows (PowerShell, no admin needed):

powershell -ExecutionPolicy ByPass -c "irm https://astral.sh/uv/install.ps1 | iex"
[System.Environment]::SetEnvironmentVariable("GOOGLE_OAUTH_CLIENT_ID", "your-client-id", "User")
[System.Environment]::SetEnvironmentVariable("GOOGLE_OAUTH_CLIENT_SECRET", "your-client-secret", "User")
[System.Environment]::SetEnvironmentVariable("USER_GOOGLE_EMAIL", "you@yourdomain.com", "User")
[System.Environment]::SetEnvironmentVariable("PORT", "8765", "User")

Close and reopen PowerShell so the new env vars take effect. Git for Windows must also be installed (Claude Code uses Git Bash internally) — get it from git-scm.com with default options.

User-scope env vars are global on Windows, so if you need a second workspace account or run multiple Claude sessions concurrently, override USER_GOOGLE_EMAIL and PORT per session before launching Claude ( $env:USER_GOOGLE_EMAIL="..."; $env:PORT="8766"; claude).

Step 3 — Add the MCP server

Run this from the directory where you want the server active (the default scope is local — the registration is tied to that working directory, not your whole user account). The server inherits GOOGLE_OAUTH_CLIENT_ID, GOOGLE_OAUTH_CLIENT_SECRET, USER_GOOGLE_EMAIL, and PORT from the shell environment Claude Code is launched in, so no -e flags are needed:

claude mcp add google-workspace -- uvx workspace-mcp --tool-tier complete

(Add --scope user if you want the server available globally in every directory; add --scope project to write it to a shared .mcp.json in the repo — but never with hard-coded credentials, since .mcp.json typically gets committed.)

On first use, a browser window opens for OAuth — sign in with the matching Google account and approve the scopes. Tokens cache locally:

macOS: ~/.google_workspace_mcp/credentials/
Windows: %USERPROFILE%\.google_workspace_mcp\credentials\

No browser prompt on subsequent startups. To update workspace-mcp: uv cache clean — Claude Code fetches the latest version on next startup. To revoke access: delete the credentials directory above, or remove the Workspace MCP app at myaccount.google.com/permissions. To rotate the Client Secret if it leaks: APIs & Services → Credentials → workspace-mcp-desktop → Reset Secret, then update env vars and re-auth.

Multi-account & multi-session port assignment (important)

The workspace-mcp server binds a local port for its OAuth callback (default 8000). Every workspace-mcp instance running concurrently on the machine needs a unique port. There are two scenarios where this matters — most people anticipate the first but not the second:

Multiple accounts on the same machine (e.g. personal + company in different folders). Each folder's .envrc (macOS/direnv) — or each session's PowerShell overrides (Windows) — needs its own PORT, otherwise both instances try to bind 8000 and the second one fails.
Multiple Claude Code sessions running at the same time, even from different directories. Per-directory env vars don't help if two sessions still resolve to the same PORT — every running Claude Code session spawns its own MCP processes, and if any two of them point at the same port, the later one's workspace-mcp won't start. Even with a single account per directory, every directory whose Claude session might run concurrently with another needs its own port.

Pick ports above 8000 (e.g. 8765, 8766, 8767, …) and assign one per account and per concurrently-running session. The setup is then one .envrc per directory plus a single claude mcp add from each. Example with two accounts on macOS:

# ~/work/personal-stuff/.envrc
export GOOGLE_OAUTH_CLIENT_ID="personal-client-id"
export GOOGLE_OAUTH_CLIENT_SECRET="personal-secret"
export USER_GOOGLE_EMAIL="you@personal.com"
export PORT="8765"

# ~/work/cloudofficer/.envrc
export GOOGLE_OAUTH_CLIENT_ID="company-client-id"
export GOOGLE_OAUTH_CLIENT_SECRET="company-secret"
export USER_GOOGLE_EMAIL="you@company.com"
export PORT="8766"

Run direnv allow once in each directory, then register the server from each (no -e flags — credentials and port come from .envrc):

cd ~/work/personal-stuff && claude mcp add google-workspace -- uvx workspace-mcp --tool-tier complete
cd ~/work/cloudofficer && claude mcp add google-workspace -- uvx workspace-mcp --tool-tier complete

The port only matters during the OAuth callback handshake; it does not need to be reachable from outside the machine. Verify everything is registered (run from inside each directory):

claude mcp list

Sample prompts once it's up:

List my unread emails
Search drive for the Q3 proposal
Create a calendar event for tomorrow at 2pm

Complementary Plugins (optional)

These official plugins from the claude-plugins-official marketplace pair well with co-dev. They're independent — install only the ones relevant to your work. The claude-plugins-official marketplace is auto-installed by Claude Code, so no marketplace add step is needed.

Plugin	What it adds	Install
`plugin-dev`	Toolkit for developing Claude Code plugins — 7 skills (hooks, MCP, structure, settings, commands, agents, skills) + validators	`claude plugin install plugin-dev@claude-plugins-official`
`claude-md-management`	Audit and revise CLAUDE.md files based on session learnings	`claude plugin install claude-md-management@claude-plugins-official`
`claude-code-setup`	Analyze a codebase and recommend Claude Code automations to add	`claude plugin install claude-code-setup@claude-plugins-official`
`skill-creator`	Guided skill creation workflow (overlaps with `plugin-dev`'s skill-development skill — pick one)	`claude plugin install skill-creator@claude-plugins-official`
`code-simplifier`	Refactor recently-modified code for clarity, DRY, and consistency	`claude plugin install code-simplifier@claude-plugins-official`
`code-review`	Lean automated PR review with confidence scoring (alternative to `co-dev`'s `/code-review-deep`, which is broader and not PR-bound)	`claude plugin install code-review@claude-plugins-official`
`security-guidance`	Proactive security hints via PreToolUse hooks (complements `code-review-deep`'s reactive audit)	`claude plugin install security-guidance@claude-plugins-official`
`frontend-design`	Opinionated patterns for production-grade frontend interfaces	`claude plugin install frontend-design@claude-plugins-official`
`superpowers`	Brainstorming and subagent-driven workflows	`claude plugin install superpowers@claude-plugins-official`
`agent-sdk-dev`	Toolkit for building apps on the Anthropic Agent SDK (a different stack from Claude Code plugins)	`claude plugin install agent-sdk-dev@claude-plugins-official`
`document-skills`	Anthropic's `xlsx`/`docx`/`pptx`/`pdf` skills — deliver `co-dev` reports (`weekly-dev-report`, `sprint-summary`, `query-db`, …) as Office/PDF files. Separate marketplace (see note).	`claude plugin marketplace add anthropics/skills` then `claude plugin install document-skills@anthropic-agent-skills`
`stripe`	Stripe's official plugin — integration skills (best practices, SDK/API upgrades) plus Stripe's own MCP + agent tools	`claude plugin install stripe@claude-plugins-official`
`vercel`	Vercel's official plugin — deployment ops via the Vercel MCP, plus `/deploy` `/env` `/status` commands and a deployment-expert agent	`claude plugin install vercel@claude-plugins-official`
`huggingface-skills`	Hugging Face's official skills — build, train, evaluate, and use open-source models, datasets, and Spaces (HF CLI, trainers, evaluation, Gradio). For ML/model work.	`claude plugin install huggingface-skills@claude-plugins-official`
`sentry`	Sentry error monitoring — access error reports, analyze stack traces, search issues, debug production errors, and set up Sentry SDKs	`claude plugin install sentry@claude-plugins-official`
`azure`	Microsoft's official Azure plugin — bundles the Azure MCP + ~28 ops skills: list resources, deploy, validate, diagnose, RBAC/Entra, and optimize costs across 50+ services	`claude plugin install azure@claude-plugins-official`
`firebase`	Google Firebase plugin — bundles the Firebase MCP to manage Firestore, auth, cloud functions, hosting, and storage. Complements `co-dev`'s `crashlytics` skill	`claude plugin install firebase@claude-plugins-official`

The stripe plugin bundles Stripe's own MCP and agent tools, so you do not need a separate claude mcp add stripe. For local webhook testing you can optionally add the Stripe CLI (used by Stripe's plugin and your app dev, not by co-dev itself): brew install stripe/stripe-cli/stripe then stripe login (stripe listen, stripe trigger).

The vercel plugin bundles the Vercel MCP, so no separate claude mcp add vercel is needed. That MCP is read-only (docs, list projects/deployments, logs); write operations (deploy, promote, env changes) run through the vercel CLI — install it too: brew install vercel-cli then vercel login.

The huggingface-skills plugin's Hub operations use the Hugging Face CLI — install it with brew install hf then hf auth login. Its training and evaluation skills additionally need a Python ML environment (torch, transformers, TRL) specific to your project.

The sentry plugin is skill-based (SDK setup + debug workflows). For CLI operations — release tracking and source-map uploads — install the Sentry CLI: brew install sentry-cli (also packaged as the sentry-cli plugin).

The azure plugin bundles the Azure MCP (@azure/mcp), so no separate claude mcp add is needed. It authenticates with your Azure CLI credentials — install it with brew install azure-cli then az login.

The firebase plugin bundles the Firebase MCP, so no separate claude mcp add is needed. Its CLI operations (deploy, emulators, security rules) use the Firebase CLI — install it with brew install firebase-cli then firebase login. It complements co-dev's crashlytics skill (crash data via BigQuery) rather than replacing it.

document-skills lives on Anthropic's own anthropic-agent-skills marketplace (not claude-plugins-official), so it needs the one-time claude plugin marketplace add step above. It is not open source — review each skill's LICENSE.txt. Optional: Claude Code can already generate these formats on demand via open-source libraries (openpyxl, python-docx, python-pptx, reportlab); the skills mainly add standardized, repeatable formatting.

Configure Environment Variables

All credentials are read from environment variables — they are never stored in the plugin. Each skill's required vars are listed in the Setup column of the Skills table below.

For switching between accounts/projects per directory, see the Multi-Account Setups section.

Multi-Account Setups (direnv)

Every account-bound MCP server in this plugin reads its credentials from environment variables — no credentials are stored in .mcp.json. This makes direnv a natural fit for switching between accounts (different AWS profiles, separate Postgres instances, multiple GitHub orgs, etc.) by setting per-directory env vars in an .envrc file.

Important caveat: Claude Code spawns MCP servers once, at startup, and they inherit the shell environment at that moment. The implication:

✓ Works: cd ~/project-a && claude — MCPs pick up project-a's env vars from .envrc. Quit, cd ~/project-b && claude — MCPs pick up project-b's env vars.
✗ Does not work: starting Claude Code in one project, then cd-ing to another mid-session. The already-running MCP servers keep the original env vars and continue talking to the original account.

Bottom line: to switch accounts, quit Claude Code and relaunch it from the target directory. direnv handles the rest.

Recommended Permissions

This plugin bundles several MCP servers. By default, Claude Code will prompt for permission each time an MCP tool is called. To auto-approve these tools, add the following entries to the permissions.allow array in your ~/.claude/settings.json:

{
  "permissions": {
    "allow": [
      "mcp__appstore__*",
      "mcp__aws__*",
      "mcp__chrome-devtools__*",
      "mcp__context7__*",
      "mcp__fetch__*",
      "mcp__gcloud__*",
      "mcp__github__*",
      "mcp__mongodb__*",
      "mcp__mysql__*",
      "mcp__playstore__*",
      "mcp__playwright__*",
      "mcp__postgres__*",
      "mcp__redis__*"
    ]
  }
}

If you also use remote MCP servers (see Configure Remote MCPs), add their patterns too:

{
  "permissions": {
    "allow": [
      "mcp__atlassian__*",
      "mcp__bigquery__*",
      "mcp__figma__*",
      "mcp__monday__*",
      "mcp__newrelic__*",
      "mcp__paypal__*"
    ]
  }
}

Note: These entries merge with your existing allow list — you don't need to replace it. Only add entries for the MCP servers you actually use.

Renamed instances (see Naming caveat for multi-tenant setups) need their own permission entry — e.g. mcp__atlassian-companya__* and mcp__atlassian-companyb__* instead of (or in addition to) the bare mcp__atlassian__*.

Usage

Skills

These skills are automatically available to Claude. The Setup column lists what you need to configure for each — env vars, one-time auth commands, or remote MCP additions. Skills with no setup work out of the box.

Skill	Description	Setup
`analyze-db`	Generate docs/db.md with database schema docs	DB-specific env vars — see `query-db` row
`appstore`	Manage App Store Connect (builds, TestFlight, reviews, IAPs)	macOS-only. `mint install zelentsov-dev/asc-mcp`. Env: `ASC_KEY_ID`, `ASC_ISSUER_ID`, `ASC_PRIVATE_KEY_PATH` (path to `.p8`). Create API key
`aws`	Manage AWS infrastructure and services	`aws configure`, or env: `AWS_PROFILE` (or) `AWS_ACCESS_KEY_ID` + `AWS_SECRET_ACCESS_KEY` + `AWS_REGION`
`code-review-deep`	Exhaustive multi-phase code audit using parallel agents	None — uses bundled `github` + `context7` MCPs
`create-issue`	Create GitHub or Jira issues with proper templates	GitHub: `GITHUB_PERSONAL_ACCESS_TOKEN` (or) `gh auth login`. Jira: atlassian remote MCP (or) `jira init`
`create-pr`	Generate PR content, open the PR, return to default branch	GitHub: `gh auth login` (or) `GITHUB_PERSONAL_ACCESS_TOKEN`
`crashlytics`	Query Firebase Crashlytics crash data from BigQuery	bigquery remote MCP (or) `gcloud auth application-default login`. Env: `BQ_PROJECT`, `BQ_CRASHLYTICS_DATASET`
`doc-tracker-coverage`	Verify a Google Doc section's items are tracked in Monday/Jira	Google Doc read: Workspace Docs MCP (or) `claude.ai Google Drive` MCP (or) published-to-web doc. monday: `claude mcp add monday` (`${MONDAY_TOKEN}`). Jira: atlassian remote MCP (or) `jira init` Optional env: `DOC_TRACKER_COVERAGE_TEAMS`
`gcloud`	Manage Google Cloud infrastructure and services	`gcloud auth login && gcloud config set project <id>`. Optional: `CLOUDSDK_ACTIVE_CONFIG_NAME` for named configs
`heroku`	Manage Heroku apps, dynos, logs, and databases	`heroku login`
`loco`	Manage Loco translation assets (create, delete, scan)	Env: `LOCO_API_KEY` (or) per-project `LOCO_API_KEY_<PROJECT>` (e.g. `LOCO_API_KEY_IOS`)
`migrate-code`	Port a codebase to another language/framework via a six-step rulebook/translate/compile/test/verify engine	None — uses the repo's own build/test commands + bundled `context7` MCP
`monday`	Manage monday.com boards, items, groups, columns, and updates	`claude mcp add monday` with a `${MONDAY_TOKEN}` Bearer header — see Configure Remote MCPs. Token from avatar → Developers → My access tokens. MCP-only, no CLI fallback
`monday-weekly-report`	Weekly project-status report from monday.com board(s)	`claude mcp add monday` with a `${MONDAY_TOKEN}` Bearer header (see `monday` above). For attribution enable dynamic API. Optional env: `MONDAY_WEEKLY_REPORT_BOARDS`, `MONDAY_WEEKLY_REPORT_TO` (required for `--send`), `MONDAY_WEEKLY_REPORT_CC`
`newrelic`	Query New Relic observability data, alerts, and logs	newrelic remote MCP (or) `newrelic profile add --name default --apiKey NRAK-... --accountId ...`
`paypal`	Manage PayPal invoices, payments, and disputes	paypal remote MCP (required — no CLI fallback)
`playstore`	Fetch, analyze, and respond to Google Play reviews	Env: `GOOGLE_APPLICATION_CREDENTIALS` (path to service-account JSON; needs Google Play Developer API + Play Console access). Requires `uv`/`uvx` on PATH.
`query-db`	Query databases using natural language via CLI	Per DB: PG (`PGHOST`, `PGPORT`, `PGUSER`, `PGPASSWORD`, `PGDATABASE`) · MySQL (`MYSQL_*`) · Mongo (`MDB_MCP_CONNECTION_STRING`) · Redis (`REDIS_URL`) · SQLite (`SQLITE_DB`) · BigQuery (`BQ_PROJECT`, `BQ_DATASETS`) · ES (`ES_URL`, `ES_API_KEY`)
`review-architecture`	Review or create docs/architecture.md	None — uses bundled `fetch` + `context7` MCPs
`review-copy`	Audit user-facing copy — microcopy, form/help text, voice & tone, i18n readiness, marketing; writes docs/copy-review.md	None — uses `git` + optional `loco` handoff for i18n keys
`review-design`	Compare UI code against Figma designs (Android, iOS, web)	figma remote MCP (required — no CLI fallback)
`review-ownership-map`	Code ownership & bus factor from git history — flags single points of failure on sensitive code	None — read-only git analysis
`review-readme`	Review or create README.md to match standards	None — uses bundled `fetch` + `context7` MCPs
`review-seo`	Audit SEO + GEO + front-end web quality (perf, Core Web Vitals, accessibility) for a codebase and/or live URL	None — uses `curl` + bundled `fetch` + `chrome-devtools` MCPs
`review-threat-model`	Repo-specific STRIDE threat model — trust boundaries, abuse paths, ISO 27001 evidence	None — read-only code analysis
`review-user-guide`	Review or create docs/user-guide.md with user documentation	None — uses bundled `fetch` + `context7` MCPs
`run-linters`	Run linters and fix any issues found	None — runs project linters already configured in the repo
`sprint-summary`	Summarize sprint items grouped by repo in ~3-day blocks	atlassian remote MCP (or) `jira init`
`verify-resolved-issues`	Audit resolved/fixed/done issues, verify fixes, close or reopen	GitHub: `gh auth login` (or) `GITHUB_PERSONAL_ACCESS_TOKEN`. Jira: atlassian remote MCP (or) `jira init`
`weekly-dev-report`	Weekly dev activity report from Jira sprint + GitHub	atlassian remote MCP (or) `jira init`. GitHub: `gh auth login`. Optional env: `WEEKLY_DEV_REPORT_TO` (required for `--send`), `WEEKLY_DEV_REPORT_CC`, `GITHUB_USERNAME_MAP`
`work-issue`	Implement GitHub/Jira issue(s) end-to-end and open a PR	GitHub: `gh auth login` (or) `GITHUB_PERSONAL_ACCESS_TOKEN`. Jira: atlassian remote MCP (or) `jira init`
`write-tests`	Generate and run tests in the repo's own framework (unit/integration/e2e); enforces 80/80 line/branch coverage	None — uses the test runner already in the repo (RSpec, pytest, Jest/Vitest, go test, dotnet, xcodebuild, gradle, PHPUnit/Pest, cargo, ctest, …)

Language Servers (LSPs)

The plugin declares LSPs for 13 languages. Each is idle until you open a matching file in your workspace — opening app.ts activates the TypeScript LSP, opening lib.go activates gopls, etc. No LSP runs in a repo that doesn't have matching file extensions.

Why they matter: an active LSP gives Claude real type information from your installed dependencies — function signatures, references, type definitions, completions — instead of inferring from source code. This is what stops Claude from hallucinating method names on third-party libraries or guessing the shape of a return type. Any skill that touches code (code-review-deep, migrate-code, run-linters, work-issue, write-tests, create-pr, the deep-analysis agents) becomes meaningfully more accurate.

Activation rules:

LSPs spawn on demand when matching file extensions exist in the workspace, and stop when Claude Code exits.
If the LSP binary is not on your PATH, that language silently falls back to source-only analysis — the rest of the plugin is unaffected.
Each LSP communicates over stdio with Claude Code's LSP host; you don't interact with them directly.
They are independent of the bundled MCP servers — different protocol, different lifecycle, different purpose. MCPs give Claude tools (e.g., "search GitHub", "query a database"). LSPs give Claude knowledge of your code.

Bundled LSPs:

Language	LSP binary	File extensions
TypeScript /JavaScript	`typescript-language-server`	`.ts`, `.tsx`, `.js`, `.jsx`, `.mts`, `.cts`, `.mjs`, `.cjs`
Python	`pyright-langserver`	`.py`, `.pyi`
Ruby	`ruby-lsp`	`.rb`, `.rake`, `.gemspec`, `.ru`
Go	`gopls`	`.go`
Bash	`bash-language-server`	`.sh`, `.bash`, `.zsh`
Rust	`rust-analyzer`	`.rs`
YAML	`yaml-language-server`	`.yml`, `.yaml` (provides GitHub Actions / k8s schema validation)
Swift	`sourcekit-lsp` (bundled with Xcode)	`.swift`
PHP	`intelephense`	`.php`
Kotlin	`kotlin-lsp` (JetBrains, pre-alpha)	`.kt`, `.kts` — JVM-only Gradle/Maven projects; KMP not yet supported
Java	`jdtls`	`.java`
C / C++ / Objective-C	`clangd`	`.c`, `.cpp`, `.cc`, `.cxx`, `.h`, `.hpp`, `.hh`, `.hxx`, `.m`, `.mm`
Perl	`perlnavigator`	`.pl`, `.pm`, `.t`

Install commands per LSP are in the macOS Quick Start section. Install only the LSPs for languages you actually write — each one adds a small startup cost only when its file types appear in a workspace.

Local Development

claude --plugin-dir /path/to/claude-code-plugin-dev

Contributing

We love your input! We want to make contributing to this project as easy and transparent as possible, whether it's:

Reporting a bug
Discussing the current state of the code
Submitting a fix
Proposing new features
Becoming a maintainer

Pull requests are the best way to propose changes to the codebase. We actively welcome your pull requests:

Fork the repo and create your branch from master.
If you've added code that should be tested, add tests. Ensure the test suite passes.
Update the documentation.
Make sure your code lints.
Issue that pull request!

When you submit code changes, your submissions are understood to be under the same License that covers the project. Feel free to contact the maintainers if that's a concern.