fd-open-data-mcp
The open-data ontology MCP — one protocol, every source.
The problem
AI agents need real-world data — stock prices, macro indicators, government statistics, SEC filings. Today every source speaks its own API, its own units, its own quirks. Agents end up hardcoding source-specific calls, and break when a source changes or goes down.
What it is
fd-open-data-mcp is a semantic concept layer over the world’s open data. Instead of calling source APIs, you ask for concepts and entities:
price.closeof Kweichow Moutai — not “akshare stock_zh_a_hist of 600519”gdp.current_usdof China — not a World Bank indicator code you had to memorize
The server resolves each concept against every source that carries it, ranks candidates by data quality, fails over when a source is down, and refreshes cached values on a schedule. One protocol, one MCP, every source.
How it fits
The server speaks MCP over streamable HTTP — plug it into Claude, Cursor, or any MCP client, or try it right now in the live demo. It is the reference implementation of the fd-open-data-protocol, and the data backbone for the rest of the org’s projects.
The full technical README — install, configuration, tool reference — lives on the project page.