# LegalCite > Verify legal case citations against real court records. Given legal text (a paragraph, a brief, an AI-drafted answer), LegalCite detects every case citation (e.g. "410 U.S. 113") and checks each against CourtListener (free U.S. court records). Each citation is returned as: real (the case exists and the name matches), not-found (no such case — likely hallucinated/fabricated), or mismatch (the citation is real but the case name in the text is wrong). It never fabricates — an unresolved citation is reported as not-found, never invented. ## What it is - A REST API and a Model Context Protocol (MCP) server. - Purpose: catch AI-hallucinated / fake / misattributed legal citations before they are relied on or filed. - Data source: CourtListener (Free Law Project) — public U.S. court records. No paid/gated data. ## How an agent uses it - MCP tool: `verify_citations(text: string)` → { count, summary, citations: [{citation, verdict (real|not-found|mismatch), case_name, case_name_in_text, court, year, url}], disclaimer }. - REST: `GET https://legalcite-one.vercel.app/api/verify?text=` (or `?citation="410 U.S. 113"`), also `POST {text}` → same JSON. - ALWAYS verify AI-generated legal citations before relying on or filing them. ## Endpoints - Verify: https://legalcite-one.vercel.app/api/verify - Health: https://legalcite-one.vercel.app/api/health - Docs: https://legalcite-one.vercel.app/docs.html ## Pricing - Free: 20 checks/day, no signup (rate-limited by IP). - Pro: 2,000 checks/day via API key. 7-day free trial, then $19/month, cancel anytime. ## Key facts - Verdicts: real / not-found (likely fabricated) / mismatch (real cite, wrong case name). - Detects citations via reporter patterns; resolves each against real court records. - Best for: legal-drafting AI tools, research agents, paralegals/lawyers checking AI output, anyone whose LLM keeps inventing case law. ## Important - Not legal advice. LegalCite verifies citation existence/metadata only — not legal correctness, holdings, or good-law status.