CourtListener

Name: CourtListener MCP Server
Availability: InStock
Author: CourtListener

Productivity14 tools

Search US case law, PACER dockets, judges and oral arguments from your AI assistant, and verify every citation in a brief against a real database. Free and nonprofit-run, with a default rate limit of 5 requests per minute that shapes how you should use it.

Verified connector

Listed by Anthropic as a partner connector in its Connectors Directory.

Connection checked by Agentman on 19 August 2026.

Anthropic states this reflects the level of review a connector received, not a security audit.

Connect CourtListener via MCP

https://mcp.courtlistener.com/

Works in any MCP-compatible client. In Agentman Studio it is one click — no config file to edit.

CourtListener Tools & Capabilities (14)

Tool names from Anthropic's directory listing. This server requires sign-in, so we could not read tool descriptions or parameter schemas.

Limits

Rate limits are tight and they bind. 5 requests per minute, 50 per hour and 125 per day by default, rolling window, all three concurrent. This is the practical ceiling on agentic use.
The service is run by a small nonprofit. Free Law Project is a 501(c)(3). Looping call_endpoint against it is not a neutral act; its terms ask you not to route around limits with multiple accounts or credential rotation.
Free Law Project states it is not your lawyer and that its services do not provide legal advice.
Free Law Project warns specifically about AI clients. Its terms state that an answer read through an AI client is that client's output, not theirs, and may summarize, paraphrase, omit or misstate the underlying record even when the record is accurate. It asks you to verify against the primary source.
No warranty on accuracy. Free Law Project states it makes no representations regarding the truthfulness, accuracy, relevancy or reliability of material on the services, which are provided "as is", and that documents may be unreliably reproduced.
No measured accuracy rate is published for the citation tooling. The mechanism is documented in source; its false-positive and false-negative rates are not. The name-similarity threshold is 0.8, so an unusual but legitimate name variant can in principle trip a warning.
RECAP is not a complete mirror of PACER. It is built from documents people have purchased and contributed, plus targeted scraping and bulk purchases. Absence from RECAP is not evidence that a filing does not exist.
The judges database is small, in Free Law Project's own description.
Citation analysis is batched at 250 unique citations per call, and job state expires after one hour of inactivity. Session state is also keyed to the credential, so a job started under an API token is invisible over OAuth.
Four tools write to your account. Alert creation and deletion and docket subscribe and unsubscribe are annotated as not read-only; the two deletes are annotated destructive. Free Law Project asks that you review what a client is about to do before authorizing them.
An API token grants full account API access, including those write tools, and does not expire or rotate. Free Law Project says to treat it like a password.
Everything a tool returns goes to your AI client, and to the company hosting it if it is hosted. Free Law Project advises thinking carefully before sending confidential, privileged or client material through a connected assistant — which includes pasting a draft brief in for citation checking.
Third-party copyright can persist in filings. Free Law Project notes that while opinions and filings are generally public domain, other court filings may contain copyrighted books or articles that retain protection.
We could not verify tool behaviour ourselves. The endpoint returned HTTP 401 to an anonymous handshake on 2026-08-19. Tool names, annotations and internal logic come from Anthropic's directory snapshot and Free Law Project's published source code, not from a live handshake.
Prompts are unconfirmed. Anthropic's snapshot lists no prompt names, and the gated endpoint blocked our own check. Free Law Project's source does contain a prompts module, but it holds server instructions for the model rather than MCP prompt definitions.
The directory listing is stale by two tools, both read-only. Expect read_document and search_document in addition to the 14 listed.

Frequently asked questions

Can the CourtListener MCP server tell me whether a case citation is real?

Yes — that is what `analyze_citations` is built for. It extracts citations with eyecite, then checks each one against CourtListener's database and returns the canonical case name, court and date. If a citation matches by volume and reporter but the case name in your text does not match the real case name, it emits a warning naming a possible hallucinated citation.

Is the CourtListener MCP server free to use?

Yes. Free Law Project is a 501(c)(3) nonprofit and CourtListener accounts are free for individuals. You need a free account to sign in, and no separate API token for the hosted server. Free Law Project offers memberships and commercial agreements that raise API limits, but the connector itself carries no charge and no paid tier is required.

What are the CourtListener API rate limits for an AI agent?

Five requests per minute, 50 per hour and 125 per day by default for authenticated users, on a rolling window, with all three throttles applying at once. Free Law Project documents that the most restrictive limit controls. For an agent that fans out many searches this is the binding constraint, and memberships or commercial agreements are the documented way to raise it.

Can the CourtListener MCP server change anything in my account?

Yes, through four tools: `create_search_alert`, `delete_search_alert`, `subscribe_to_docket_alert` and `unsubscribe_from_docket_alert`. All four are annotated as not read-only in Free Law Project's source, and the two delete tools are annotated destructive. Everything else reads. The generic `call_endpoint` tool cannot write — it only ever issues list requests.

How much US case law does CourtListener actually cover?

Free Law Project states its database holds over nine million legal decisions from more than two thousand courts, and that it covers more than 99.9 percent of all precedential case law published in the United States. Its archives reach back to the 1650s. The RECAP archive separately holds hundreds of millions of federal docket entries sourced from PACER.

Does the directory list every tool the CourtListener MCP server has?

No. Anthropic's 2026-08-16 snapshot names 14 tools; Free Law Project's own source code registers 16. The two missing are `read_document` and `search_document`, both read-only, which page through opinion and RECAP document text and let an agent grep a document without loading all of it. The server's own instructions tell models to prefer them.

Can I reuse what the CourtListener MCP server returns?

US judicial opinions are generally public domain, and Free Law Project says so directly. It also warns that other court filings may contain third-party copyrighted works that retain protection. Its terms ask that if you republish its data you not present it as though Free Law Project produced or endorsed an AI-generated analysis of it. That is their stated position, not legal advice.

Sources

CourtListener MCP README — https://raw.githubusercontent.com/freelawproject/courtlistener-api-client/main/MCP_README.md (fetched 2026-08-19 as the raw file, 16KB; the source for the 14 documented tool names and descriptions, both auth schemes, session-state behaviour, the 250-citation batch limit, and the health URL) · retrieved 2026-08-19
Free Law Project MCP tool registry — https://github.com/freelawproject/courtlistener-api-client/blob/main/courtlistener/mcp/tools/__init__.py (retrieved 2026-08-19; the source for the 16-tool registry and its ordering, which is how the two omitted tools were identified) · retrieved 2026-08-19
Free Law Project citation analysis source — https://github.com/freelawproject/courtlistener-api-client/blob/main/courtlistener/mcp/tools/analyze_citations_tool.py (retrieved 2026-08-19; the two-stage extract-then-verify design and the case-name cross-check that flags a possible hallucinated citation) · retrieved 2026-08-19
Free Law Project citation utilities — https://github.com/freelawproject/courtlistener-api-client/blob/main/courtlistener/mcp/tools/citation_utils.py (retrieved 2026-08-19; the 0.8 name-similarity threshold, the 250-citation-per-request constant, the similarity computation, and the slip-opinion handling) · retrieved 2026-08-19
Free Law Project dispatcher source — https://github.com/freelawproject/courtlistener-api-client/blob/main/courtlistener/mcp/tools/call_endpoint_tool.py (retrieved 2026-08-19; the read-only annotation and the list-only implementation that establishes the dispatcher cannot write) · retrieved 2026-08-19
Free Law Project endpoint registry — https://github.com/freelawproject/courtlistener-api-client/blob/main/courtlistener/models/endpoints/__init__.py (retrieved 2026-08-19; the 48 registered REST endpoints reachable through the dispatcher) · retrieved 2026-08-19
Free Law Project server instructions — https://github.com/freelawproject/courtlistener-api-client/blob/main/courtlistener/mcp/prompts.py (retrieved 2026-08-19; the four search collections, the direction to prefer the two document tools, and the field-name caveats) · retrieved 2026-08-19
CourtListener REST API documentation — https://wiki.free.law/c/courtlistener/help/api/rest/v4/rest-api-v47 (fetched 2026-08-19 after a 301 from courtlistener.com/help/api/rest/; the 5/50/125 rate limits, the rolling-window and concurrent-throttle rules, the single-account rule, 3,359 jurisdictions, and Free Law Project's own framing of citation lookup as an anti-hallucination guardrail) · retrieved 2026-08-19
CourtListener case law coverage — https://wiki.free.law/c/courtlistener/help/data-coverage/case-law (fetched 2026-08-19; 9M+ decisions, 2,000+ courts, the 99.9% precedential claim, and the Harvard, vLex and Caselaw Access Project sources) · retrieved 2026-08-19
CourtListener federal cases and filings coverage — https://wiki.free.law/c/courtlistener/help/data-coverage/federal-cases-and-filings (fetched 2026-08-19; RECAP's data sources, ~30,000 extension users, and roughly 100,000 new docket entries on a given day) · retrieved 2026-08-19
CourtListener data coverage overview — https://wiki.free.law/c/courtlistener/help/data-coverage (fetched 2026-08-19; archives to the 1650s and the "small database of judges" description) · retrieved 2026-08-19
CourtListener terms of service and policies — https://wiki.free.law/c/terms/courtlistener/courtlistenercom-terms-of-service-and-policies (fetched 2026-08-19; the Automated and Agentic Access section, the AI-client accuracy warning, the public-domain and third-party-copyright position, the warranty disclaimer, and the AI-provider privacy section; terms last modified 2026-08-05) · retrieved 2026-08-19
Live RFC 9728 protected-resource descriptor — https://mcp.courtlistener.com/.well-known/oauth-protected-resource (fetched 2026-08-19, HTTP 200, 177 bytes; scopes openid and api, one authorization server, header-only bearer methods) · retrieved 2026-08-19
Live RFC 8414 authorization-server metadata — https://www.courtlistener.com/.well-known/oauth-authorization-server (fetched 2026-08-19, HTTP 200, 871 bytes; issuer, dynamic client registration, PKCE S256, and the same two scopes). The same document on the MCP host returns 404 · retrieved 2026-08-19
Live anonymous probe — 2026-08-19: initialize to the bare endpoint returns HTTP 401 with www-authenticate: Bearer error="invalid_token" and a resource_metadata pointer at the descriptor above. No tool listing was obtainable · retrieved 2026-08-19
Doc discovery — courtlistener.com/robots.txt asks crawlers to make contact and points at the REST API and bulk data instead, disallowing the retired v1–v3 API paths; llms.txt returns 404. Neither host returned a Content-Signal header (both fetched 2026-08-19) · retrieved 2026-08-19
Free Law Project data services — https://free.law/data-consulting/ (fetched 2026-08-19; the CC BY-ND 4.0 site content licence and the paid data-services route for bulk needs) · retrieved 2026-08-19
Anthropic connector directory — https://claude.ai/directory/58d5cff1-20b3-4bb8-9c51-9ac3e6bd0dac (snapshot 2026-08-16; permissions label "Read and write", 14 tool names, no prompt names, partner tier) · retrieved 2026-08-16