@pipeworx/court-listener

Connect: https://gateway.pipeworx.io/court-listener/mcp · Install: one-click buttons

Tools: 10

US court opinions from Free Law Project’s CourtListener: 8.3 million opinions across 3,361 courts, going back to 1761. The largest free, structured database of American case law.

Most of it is state law. That surprises people, and it is the point:

Federal appellate law is easy to find elsewhere. Six million state opinions, assembled and addressable, are not.

Auth

None for case lookup and opinion text — no key, no signup, no rate limit.

search_opinions and search_dockets search upstream by topic and are subject to CourtListener’s throttling. A free token from Free Law Project lifts it.

Tools

For litigation documents the path is search_dockets → list_docket_filings → get_docket_filing. Every filing tool is keyed by docket_id, and only search_dockets produces one — a PACER case number will not resolve.

Opinions and filings answer different questions: an opinion is what a court decided, a filing is what the parties filed.

search_opinions returns one cases shape whichever path served it (mirror or live), and the most-cited case inlines an excerpt of its lead opinion — the opening, its holding language, and the query match — so “what did the court hold” is answerable from the search result alone; get_opinion reads the full text.

The usual path is find_case → list_case_opinions → get_opinion. Results from find_case and lookup_citation are ranked by citation count, because among cases sharing a name the one later courts actually cite is nearly always the one being asked about — “roe wade” returns the 1973 Supreme Court decision (5,581 citations) ahead of the 1970 district case (53).

Abbreviated and copy-pasted case names

Name matching is AND over every token, so one token the corpus does not carry sinks the whole lookup. That used to mean find_case returned nothing for Bell Atl. Corp. v. Twombly — the corpus stores “Bell Atlantic Corp.”, and atl does not stem to atlant — and nothing for a name pasted out of a brief with its scaffold attached (TABLE OF AUTHORITIES Bowers v. Hardwick).

When the name as given matches nothing, find_case now retries once with Bluebook abbreviations, citation scaffold and corporate suffixes dropped (never expanded or guessed — under AND semantics dropping a token can only widen). A result found that way is labelled:

{ "matched": "relaxed", "matched_query": "Bell v Twombly", "note": "…" }

Treat a matched: "relaxed" hit as a best-effort resolution of a messy citation and check the returned case_name is the case you meant. An exact match never carries the field, and a name that matches on the first pass still costs exactly one query.

Coverage

Keyless: opinion full text, case names, filing dates, judges, precedential status, citation counts, reporter citations.

Upstream, throttled: full-text topic search, and all three filing tools. Finding a case by topic can fail in a way that finding it by name or citation cannot.

RECAP filings (PACER documents) — keyless, via search only. ~9.7 million documents. The REST endpoints you would reach for, /recap-documents/ and /docket-entries/, both answer “Authentication credentials were not provided”; /search/?type=rd needs no token and is the path the filing tools use. Scoping goes through the q field query (q=docket_id:71886217) — passing ?docket_id= as a parameter is silently ignored and returns the whole 61.7M-document corpus rather than an error.

Caveats worth passing to a user

• A filing without a PDF is still a real filing. RECAP holds docket text for far more entries than it holds documents, because a PDF only exists once somebody has paid PACER for it. Each filing reports pdf_available; on a typical docket most entries are metadata-only. Read a missing PDF as “nobody has bought this one”, never as “this was not filed”.
• Filing tools spend a scarce budget. They go upstream to CourtListener’s /search/, throttled to roughly 125 requests a day without a token — the same throttle the opinion mirror exists to avoid. There is no mirror for RECAP. When it is exhausted the tools say so explicitly rather than returning an empty docket; opinion tools are unaffected.
• Recency. Responses carry a snapshot_date; opinions filed after it fall through to upstream search.
• 14.2% of the corpus is OCR of scanned volumes and reads accordingly. Responses set extracted_by_ocr so garbled text is attributable to the scan rather than to us.
• date_created is not a filing date. It is when the record entered CourtListener — a 1946 opinion can read 2020. date_filed, joined from the cluster, is when the court actually ruled.
• Citation counts are within CourtListener’s corpus, not a universal measure of importance.

Attribution

Data is CourtListener / Free Law Project, published under the Public Domain Mark. Free Law Project funds and maintains the corpus. If this pack is useful to you, https://free.law/donate/ is where that goes.

Tools

• search_opinions — US court opinions matching a keyword search, from CourtListener’s 8.3 million opinions across state and federal courts.
• search_dockets — US federal court dockets matching a keyword search, from the PACER/RECAP archive.
• search_case_law — US court opinions matching a full-text subject search, from CourtListener’s 8.3 million opinions.
• find_case — A US court case’s citations, filing date and judges, found by case name, from CourtListener.
• lookup_citation — The case a US reporter citation (e.g. “410 U.S. 113”) refers to, from CourtListener.
• list_case_opinions — Every opinion filed in one US court case — majority, concurrences, dissents — from CourtListener.
• get_opinion — The full text of one US court opinion by its CourtListener id.
• list_docket_filings — Every filing on one US federal court docket, in filing order, from the PACER/RECAP archive.
• get_docket_filing — One filing from a US federal court docket, by its RECAP document id.
• search_court_filings — US federal court filings — complaints, motions, orders — matching a text search, from PACER/RECAP.