---

title: GraphPulse emoji: 🐺 colorFrom: indigo colorTo: gray sdk: docker app_port: 7860 pinned: false

LegacyGraph-MCP: Agentic C++ Modernization 🏗️

⚡ The Problem

Legacy C++ codebases are too large for LLM context windows, leading to:

1. Lost Context – spaghetti code cannot be processed in one pass.
2. Hallucinations – agents refactor without full dependency awareness.
3. Parsing Gaps – regex‑based parsers miss macros and templates.

🛠️ The Solution

LegacyGraph‑MCP is a Model Context Protocol (MCP) server that exposes a C++ codebase as a knowledge graph. Agents query the graph instead of raw text:

"Agent: Which functions call calculate_risk()?" "MCP: process_loan() and assess_credit()"

Core Features

• Accurate AST parsing via tree‑sitter – 100 % C++ syntax coverage.
• Graph‑RAG – detects circular dependencies before refactoring.
• Hybrid deployment – local (stdio) or cloud (HTTP) with a single codebase.
• Universal MCP client support – works with Claude Desktop, DeepSeek‑Coder, etc.
• Omni‑ingestion – clone repos, apply patches, upload raw files, or scan local directories.
• Token‑safe visualisation – Mermaid diagrams returned inline (cloud) or saved to disk (local).

---

📊 Performance (LLVM – 67 k files)

LegacyGraph‑MCP now uses JSON for cache files (.legacygraph.json) instead of unsafe pickle, and supports a constrained‑resource benchmark profile.

Metric	Sequential (v0.1)	Distributed (v0.3)	⚡ Improvement
CPU Utilisation	~6 % (1 core)	~80 % dynamic cores	Adaptive
AST Build Time	5 h	10 min	≈30×
Throughput	3.8 files/s	113 files/s	≈30×
Incremental Updates	N/A	< 1 s	Instant

---

📐 Architecture

src/
├── __init__.py            # package root
├── __main__.py            # CLI entry point (python -m src)
├── server.py              # MCP registration & server card
│
├── core/                  # business logic
│   ├── graph.py           # NetworkX dependency graph model
│   └── parser.py          # tree‑sitter C++ parser
│
├── tools/                 # MCP tool functions
│   ├── analysis.py        # analyze_codebase (ingestion)
│   ├── queries.py         # get_callers, get_callees, detect_cycles …
│   └── export.py          # Mermaid graph generation & export
│
└── utils/                 # cross‑cutting infrastructure
    ├── config.py          # MCP_MODE, CPP_EXTENSIONS
    ├── logger.py          # centralized logging
    ├── services.py        # singletons (graph, parser)
    └── helpers.py         # git clone, directory scanning, Mermaid builder

graph LR
    A[AI Agent] -->|JSON‑RPC| B[MCP Server]
    B -->|Parse| C[tree‑sitter]
    B -->|Query| D[NetworkX Graph]
    C -->|AST| D
    D -->|Cycles/Deps| B
    E{MCP_MODE} -->|local| F[stdio + Disk]
    E -->|cloud| G[HTTP + /tmp/ Clone]

📖 See ARCHITECTURE.md for detailed component diagrams.

---

👨‍💻 Developer Guide (Run & Develop)

This section helps developers run the code directly, develop the server, and test it locally.

1. Prerequisites and Installation

Ensure you have Python 3.11+ and poetry installed.

git clone https://github.com/RohitYadav34980/LegacyGraph-MCP.git
cd LegacyGraph-MCP

# Install dependencies using Poetry
pip install poetry
poetry install

2. Running the Server Locally

To develop or verify the MCP server running natively:

Standard Local Execution (stdio)

python -m src --mode local

Cloud Simulation Mode (HTTP / SSE)

Test the cloud endpoints locally without Docker:

python -m src --mode cloud --transport http --path /mcp --host 127.0.0.1 --port 7860

(--transport streamable-http is accepted as an alias for http; sse remains available for legacy clients.) This forces the server to bind to localhost:7860, mimicking the Hugging Face Spaces environment.

3. Verify Server State

Test if your local setup works perfectly:

# Run the internal validation script
python tools/verifier.py

Expected output: 100 % accuracy on dependency detection.

---

☁️ Cloud Deployment (Hugging Face Spaces)

This project is optimized for Hugging Face Spaces using the Docker SDK.

1. Create a Space

Create a new Space on huggingface.co and select Docker as the SDK.

2. Connect & Push

The easiest way is to push your existing local repository to the Space:

# Add the Hugging Face Space as a remote
git remote add hf https://huggingface.co/spaces/Rohitadav/GraphPulse

# Push your code (requires an HF Access Token)
git push hf main --force

3. Verify

Once pushed, the Space will automatically build the Dockerfile and start the server on port 7860. You can then consume this server via SSE in any MCP-compatible client.

---

🔌 Installing in Your MCP Client

Connect your AI agent to the Graph engine using these configurations.

Option 1: Connect via Smithery (For Remote / Cloud)

If you've deployed to Hugging Face or another remote hosting, or just want to use the published Remote MCP setup, use Smithery. You can view the server directly on Smithery.

npx -y @smithery/cli@latest mcp add labsofuniverse/legacy-mcp-analyzer --client claude-code

Option 2: Local / Claude Desktop Configuration

If you want to plug your local development environment directly into Claude Desktop, update your configuration file.

• macOS: ~/Library/Application Support/Claude/claude_desktop_config.json
• Windows: %APPDATA%\Claude\claude_desktop_config.json

Add the following configuration, ensuring you replace the cwd paths with the absolute path where you cloned this repository.

{
  "mcpServers": {
    "legacy-mcp-analyzer": {
      "command": "python",
      "args": [
        "-m", 
        "src", 
        "--mode", 
        "local"
      ],
      "cwd": "C:\\path\\to\\your\\LegacyGraph-MCP"
    }
  }
}

Note: Make sure to escape backslashes on Windows (e.g. C:\\Users\\...), or use forward slashes on macOS/Linux.

---

🔧 MCP Tools

Tool	Description	Mode
analyze_codebase	Unified ingestion (repo URL, patch, raw files, or local dir)	Both
get_file_functions	List functions defined in a specific source file	Both
get_file_coupling	Cross‑file coupling report (file A → file B)	Both
get_callers	Find upstream dependencies	Both
get_callees	Find downstream dependencies	Both
detect_cycles	Identify circular dependencies	Both
get_orphan_functions	Find unused code	Both
generate_mermaid_graph	Return Mermaid diagram inline (token‑safe)	Both
export_ide_graph	Save Mermaid .md file to local disk	Local only

---

🧪 Testing

# Unit + integration tests (≈30 cases)
python -m pytest tests/ -v

# End‑to‑end verifier against sample legacy project
python tools/verifier.py

Current Accuracy: 100 % (all dependencies, cycles, and orphan detection verified).

---

📚 Documentation

Review the complete documentation suite for deep-dives into the architecture and usage:

Document	Description
PROJECT_MANUAL.md	In‑depth guide, API reference, deployment modes
ARCHITECTURE.md	Detailed architecture, data flows, component diagrams
CONTRIBUTING.md	Development standards, commit protocol, PR process
CHANGELOG.md	Version history and release notes

---

🤝 Contributing

1. Fork & clone the repo.
2. Create a feature branch: git checkout -b feature/your-feature.
3. Follow strict mypy typing, Google‑style docstrings, conventional commits.
4. Run pytest – ensure all tests pass.
5. Submit a PR.

---

🙏 Acknowledgments

Built with:

• tree‑sitter
• NetworkX
• MCP
• FastMCP

---

Made with ❤️