mcp-openidea-search

An MCP server that exposes Open Idea's federated search — across 10 open-knowledge providers (OpenAlex, arXiv, Zenodo, Software Heritage, GitHub, HuggingFace, YouTube, OSHWA, Wikifactory, generic hardware) — to MCP-compatible clients (Claude Desktop, Claude Code, Cursor, the MCP Inspector, etc.).

This package is a thin client over the Open Idea v1 HTTP API. All the federation, deduplication, normalization, and caching happen on Open Idea's side. This repo is fully standalone — no dependency on the Open Idea codebase.

Install

Requires Python 3.10+.

# from PyPI (once published)
pipx install mcp-openidea-search

# or directly from source
pip install -e .

Related MCP server: Academic MCP Server

Configure

The server reads two environment variables:

Run

stdio (Claude Desktop / Claude Code)

Add to ~/Library/Application Support/Claude/claude_desktop_config.json (macOS) or the equivalent on your OS:

{
  "mcpServers": {
    "openidea": {
      "command": "mcp-openidea-search",
      "args": ["stdio"],
      "env": {
        "OPENIDEA_API_URL": "https://openidea.app",
        "OPENIDEA_API_KEY": "oi_live_xxxxxxxxxxxxxxxxxxxxxxxxxxxx"
      }
    }
  }
}

For Claude Code, add to ~/.mcp.json:

{
  "mcpServers": {
    "openidea": {
      "command": "mcp-openidea-search",
      "args": ["stdio"],
      "env": {
        "OPENIDEA_API_URL": "https://openidea.app",
        "OPENIDEA_API_KEY": "oi_live_xxxxxxxxxxxxxxxxxxxxxxxxxxxx"
      }
    }
  }
}

Restart your client and ask: "use the openidea MCP to search for transformer architectures".

HTTP/SSE (remote)

OPENIDEA_API_URL=https://openidea.app \
OPENIDEA_API_KEY=oi_live_xxx \
mcp-openidea-search http --host 0.0.0.0 --port 8000

Verify with the MCP Inspector:

npx @modelcontextprotocol/inspector http://localhost:8000/sse

Tools and resources

Tools

Each tool takes query: string and optional limit: integer (default 20). Returns a JSON payload { results: Resource[], total: number, … }.

Resources

Single items addressable by URI:

openidea://openalex:W2741809807
openidea://github:vercel/next.js

Listing resources returns an empty array — they're addressable but not enumerable. Use the search tools to discover ids, then read_resource to fetch the canonical record.

Develop

python3 -m venv .venv
source .venv/bin/activate
pip install -e ".[dev]"
pytest -q

Troubleshooting

License

MIT.