KnowJudges MCP

Source-linked U.S. judge, court, legal-AI case, and immigration-outcome research through a read-only remote MCP server.

• Website: https://knowjudges.com

• Public MCP documentation: https://knowjudges.com/mcp

• Endpoint: https://mcp.knowjudges.com/mcp

• Transport: Streamable HTTP

• Authentication: no token required for bounded public access

Current contract

The deployed server advertises 10 tools. This reference was checked against live tools/list on October 12, 2026. Discover the deployed inputSchema and outputSchema rather than assuming a client’s cached catalog is current. All tools return structured, source-linked public results and advertise readOnlyHint: true, destructiveHint: false, and openWorldHint: false. They cannot file documents, email judges, or change records.

Related MCP server: AI Incident Law

Connect

Add the endpoint in your client's remote MCP settings. Configuration names vary; a URL-only setup is sufficient where supported.

{
  "mcpServers": {
    "know-your-judge": {
      "url": "https://mcp.knowjudges.com/mcp"
    }
  }
}

Tools and inputs

Optional inputs are omitted when not needed. Dates use YYYY-MM-DD. State/jurisdiction codes in schemas that require two letters are case-insensitive.

For asylum analytics, percentages use grants plus denials; procedural outcomes are reported separately. Samples below 35 decisions are labeled rather than hidden. Zero decisions mean no rate, not 0%. Keep availability, message, denominator/geography definitions, source, period coverage, and limitations. Historical rates neither predict an individual outcome nor imply that a person can choose a judge.

Canonical public URL chaining

Use URLs returned by discovery, preserving their exact jurisdiction-qualified form. Do not construct URLs from names or use bare slugs, internal keys, or IDs.

Judge profile, biography, or practice

find_judges({"query":"Amy Totenberg","state":"GA","limit":3})
→ select the matching results[i].profile_url
→ get_judge_profile({"judge_url": selectedProfileUrl})        // identity/assignments
→ get_judge_biography({"judge_url": selectedProfileUrl})      // biography question
→ get_courtroom_presentation_rules({"judge_url": selectedProfileUrl}) // practice question

The follow-ups are alternatives selected by the question, not a requirement to call all three. profile_url in a discovery result becomes judge_url in these tool inputs.

Court information and recorded changes

find_court({"query":"U.S. District Court for the Northern District of Georgia","state":"GA"})
→ select the matching results[i].court_url
→ get_court({"court_url": selectedCourtUrl,"limit":5})
→ get_recent_rule_changes({"since":"2026-10-01","court_url": selectedCourtUrl})

A judge-specific change lookup instead uses judge_url from judge discovery. An empty court roster means no indexed current matches; a null count means roster data is unavailable. Neither establishes the actual court has no judges. Retain location-specific contact notes rather than treating one courthouse's address or phone as courtwide.

Legal-AI cases

{
  "name": "find_ai_abuse_cases",
  "arguments": {
    "query": "AI hallucinations fabricated legal citations",
    "response": "Monetary Sanctions",
    "limit": 5
  }
}

Report the recorded outcome and preserve the returned decision-link status; inclusion in the dataset alone does not establish a judicial finding that AI was used.

Immigration comparisons

compare_immigration_analytics({"entity":"judge","nationality":"Brazil","state":"CA","year":2025,"order":"ascending","limit":3})
→ lowest recorded asylum denial rates for the available filtered period
compare_immigration_analytics({"entity":"judge","nationality":"Brazil","state":"CA","year":2025,"order":"descending","limit":3})
→ highest recorded rates; retain counts, small-sample notes, and coverage
→ get_immigration_analytics({"entity":"judge","profile_url": returnedProfileUrl,"nationality":"Brazil","state":"CA","year":2025})

Use an available, non-null returned public URL for the follow-up. These examples illustrate inputs; they do not guarantee data for a particular nationality/year.

Bounds, quotas, and anti-enumeration

• Searches, court rosters, and immigration comparisons return at most 10 records. Recent changes return at most 25. Practice-topic filters accept at most 10 topics.

• No pagination, cursor, offset, batch enumeration, or bulk export is provided. Empty, wildcard, one-character, jurisdiction-only judge/court searches, and requests for all judges/courts are rejected. Use identifying name/location text plus narrowing filters.

• Sequential prefix/identifier harvesting is throttled. Daily unique judge/court budgets count a previously accessed identity only once; changing conversation sessions does not reset an OpenAI subject's quota.

• Generic anonymous callers use the IP fallback: 20 tool requests/hour, 100/day, and 50 newly accessed judge/court identities/day.

• OpenAI callers supplying _meta["openai/subject"] use an anonymous subject bucket: 60 tool requests/hour, 300/day, and 100 new identities/day. openai/session is not a durable quota key.

• Recognized Claude requests use session context when available for short-term control; no durable per-user identifier is inferred. Without a session, aggregate network controls apply. Session allowances are 60/hour, 300/day, and 100 new identities/day.

• A separate recognized-provider IP/network backstop allows 1000 requests/hour, 10000/day, and 1000 new identities/day. Ordinary networks have a 200/hour, 1000/day backstop and a 100-new-identity/day network budget. These are aggregate ceilings, not individual entitlements.

• MCP initialization/discovery/control requests count toward the network backstop, not the normal caller tool quota. REST uses the shared enforcement implementation and anonymous fallback when no MCP caller context exists.

• Rate/harvesting limits can return HTTP 429 or a bounded MCP tool failure. Respect retry information; do not rotate identities to bypass controls. Invalid inputs and missing data are different states.

For datasets or recurring/bulk use, see https://knowjudges.com/data-licensing.

Sources and verification

Keep returned KnowJudges URLs, source/official URLs, source titles, excerpts, scope/applicability, effective/revision dates, and last-checked dates where present. Judge-specific, courtwide, statewide, and nationally applicable federal/EOIR guidance are different authority scopes. Broader rules are not personal preferences. Missing documented practice findings establish neither permission nor prohibition.

last_updated describes publication, while last_verified or source_checked_at describes an evidence check. A recent publication does not mean every source was checked that day. Preserve supported year-only biography dates and factual uncertainty.

AI-case links may be verified decision destinations or explicitly labeled source fallbacks. Preserve decision_link_status, dataset_source, and summary_note; a fallback URL must not be described as a verified official court decision. Unknown ruling counts are unavailable, not zero. Check the linked source before relying on a requirement; case-specific orders may control.

Links

• REST documentation: https://knowjudges.com/api

• Public profiles and courts: https://knowjudges.com/kyj

• Terms: https://knowjudges.com/terms

• Privacy: https://knowjudges.com/privacy

• Directory listing: https://glama.ai/mcp/connectors/com.knowjudges/knowjudges

Third-party listings and client tool catalogs may lag a release; live tools/list is authoritative.