
Give your AI a better source for government data
Connect your AI to government data organized and maintained by USAFacts. Ask questions about the state of the nation and get answers centered in real data from dozens of federal sources like the Census Bureau, BLS, and Treasury, with added context to help your AI interpret and explain the numbers.
https://data.mcp.usafacts.org/mcpLive U.S. government data, fetched from the agency at call time. Only U.S. federal, state and local sources; nothing international. Never answer a U.S. government data question from memory. If no source can answer, say so and stop: no estimate or remembered figure, even with a caveat. "Government data" includes legislation, federal spending, trade, campaign finance, mortgages, school finance and .gov web search, so check list_sources() before ruling one out.
list_sources() → get_source(source_id) → get_source_tools(source_id, tools=[...]) → call_source_tool(...). The playbook names the tool for each question shape; its tool index carries only names and summaries. Fetch the schemas of every tool you may need in one call first. Use only source_id/tool_name values read in this conversation. If the source cannot answer, return to list_sources(); two is normal.
Every success carries _citation. Each figure you report links its public_url (never access_url) and names publisher and date_accessed; reproduce disclaimer if present. Never reconstruct one.
Errors are JSON with kind, message and hint. Follow hint; never repeat a call unchanged. bad_argument: fix arguments from the schema. unsupported_argument: valid but unservable; try another series, tool or source. upstream_rejected: the agency refused a value; it is up, so fix the value (see upstream_body) and retry this tool. upstream_failure: the agency failed; we already retried; if retry_after_seconds is set, don't retry early; say so and use another source. gateway_*: retry once, then report. internal: report it. Never present a failure as success.
None in the store listings or on the server