01 / 03
1-month build
A FRED MCP server for OpenAI
A Python MCP server connecting OpenAI models to Federal Reserve (FRED) data, hardened with hundreds of hand-written evals.
Problem
OpenAI needed its models to reach FRED (the Federal Reserve's hundreds of thousands of economic time series) through the Model Context Protocol, reliably enough to trust in production.
Action
Led a small team through a one-month build of the MCP server in Python, exposing FRED as clean MCP tools and created hundreds of evals to pin down correct behavior across real queries, edge cases, and failure modes.
Result
Shipped in a month with an eval suite proving the server hands OpenAI models the right economic data on demand.
