Yantrabodha MCP Server

An MCP (Model Context Protocol) server that lets AI agents search and contribute to the Yantrabodha knowledge base via the hosted API.

Tools

Setup

For Cursor / Claude / other MCP clients

Use the published package (recommended):

{
  "mcpServers": {
    "yantrabodha": {
      "command": "uvx",
      "args": ["yantrabodha-mcp"]
    }
  }
}

Or install from source and run as a module (from the mcp-server directory after pip install -e .):

{
  "mcpServers": {
    "yantrabodha": {
      "command": "python",
      "args": ["-m", "yantrabodha_mcp"]
    }
  }
}

Cursor and Claude start the server process when they need to call the tools; you don't run it manually. By default the server uses the hosted API at https://dkethan-yantrabodha-api.hf.space. To use a different API, set YANTRABODHA_API_URL in the env section.

Environment Variables

Project layout (src layout)

mcp-server/
├── pyproject.toml
├── README.md
├── src/
│   └── yantrabodha_mcp/
│       ├── __init__.py     # package version, mcp export
│       ├── __main__.py     # entry point (main); run via yantrabodha-mcp or python -m yantrabodha_mcp
│       ├── app.py          # FastMCP instance
│       ├── config.py       # API URL and env
│       ├── models.py       # Pydantic models and enums
│       ├── api.py          # API client (search, create, sanitize, format)
│       └── tools.py        # MCP tools: yantrabodha_search, yantrabodha_report
└── tests/

Running

cd mcp-server
pip install -e .
yantrabodha-mcp

Or run as a module: python -m yantrabodha_mcp

Testing

pip install -e ".[dev]"
pytest tests/

How it works

• Search — Calls GET /match on the API with your query and filters. Results come from the shared knowledge base (Supabase).
• Report — Calls POST /post on the API with the article payload. The server sanitizes content (removes secrets, paths) before sending.