Crs Reports MCP server

@pipeworx/crs-reports

Congressional Research Service reports — Congress's own nonpartisan analysis of statutes, federal programmes and policy options — searchable by subject and readable in full, with each report's authority, citations, licence and age stated on every response.

Part of Pipeworx — an MCP gateway connecting AI agents to 1576+ live data sources.

Tools

— ranked search over report titles and the summaries CRS publishes with them. Returns titles, report numbers, CRS subject topics, authors, publication dates and a summary snippet. Ranks titles and summaries, not report bodies — that boundary is stated in every response as searched, because a miss otherwise reads as "CRS has never written about this".

• crssearch(query?, topic?, productclass?, author?, publishedsince?, publishedbefore?, limit?)

number (R43255, IF13006, LSB10123). Returns the complete current-version text plus authors, topics, the bills and public laws it cites, and its age.

• crsreport(reportid, includetext?, maxchars?) — one report in full by its

revised lately, newest first. The change feed, for tracking the analysis that follows a bill or a rule.

• crsrecent(days?, topic?, productclass?, limit?) — what CRS has published or

plus analysts and product families. Call it before filtering: CRS writes U.S. Trade Policy, and a guessed Trade matches nothing.

• crs_topics(facet?, limit?) — CRS's own subject taxonomy with per-term counts,

Auth

On the Pipeworx gateway: keyless. The gateway injects the corpus — the report index and the extracted report text — so a gateway call needs nothing from the caller.

Standalone (npm i @pipeworx/mcp-crs-reports): bring an api.data.gov key. A standalone install has neither half of the corpus, so the pack falls back to the Library of Congress's own API and needs a key for it. Pass it as apiKey on any tool call; keys are free at and one key works across every api.data.gov-fronted API. With no corpus and no key every tool returns {found: false, reason: "corpusunavailable"} and says so.

The two modes are labelled, because they are not the same answer

Every response carries mode: corpus (hosted) or live_api (standalone with a key). What live mode cannot do — stated on the response rather than silently degraded:

The refusals are deliberate. An unfiltered list returned under a filtered question is the wrong answer wearing the right label, and a 3,700-character summary sitting in the field where a 61,000-character report belongs is the one substitution a caller would never notice.

If the corpus fails mid-call and the caller supplied apiKey, the answer comes from the live API instead and carries degradedfrom_corpus saying so.

Data sources

each report: title, publication date, current version, authors, CRS subject topics, related bills and public laws, and CRS's own summary. api.data.gov fronted.

• — the Library of Congress's record for

points at; the text comes from the current version of each.

• — the report PDFs the API

What the data does and does not say

Shortened. The full README is on GitHub.