One tool per endpoint fills a model's context with definitions it will never call. Search-then-call keeps that near zero, and storing results as SQLite turns the follow-up question into a query rather than another fetch.
An MCP server over a full financial-data REST API. Rather than one tool per endpoint it gives the model search, call and query: find the endpoint, call it, optionally store the result as an in-memory SQLite table, then run SQL against that. The endpoint index is built at startup from the API's llms.txt.
- Endpoints and built-in functions found by natural-language query, at three levels of detail
- Any REST endpoint called, with paginated responses carrying a next-page hint
- Results stored as an in-memory SQLite table for later querying
- SQL run over stored tables, including SHOW TABLES, DESCRIBE, CTEs and window functions
- Black-Scholes pricing and greeks applied to results: bs_price, bs_delta, bs_gamma, bs_theta, bs_vega, bs_rho
- Return and risk functions: simple_return, log_return, cumulative_return, sharpe_ratio, sortino_ratio
- Moving averages: sma and ema
An API key for the data provider. The project is marked experimental and subject to breaking changes, so pin the version if you build on it.
One command plus a key — uv tool install "mcp_massive @ git+https://github.com/massive-com/mcp_massive@v0.10.0", then supply credentials