Skip to content

Legal Citation Audit MCP Pack

Sourceright can be used as a narrow legal citation audit and enrichment connector. The useful boundary is deliberately smaller than a general legal research assistant:

  • extract legal citations from a draft, memo, brief section, chronology, or demand letter;
  • identify courts, jurisdictions, dates, provider candidates, and citation model issues;
  • enrich records with public-source provenance where an adapter exists;
  • return confidence, conflicts, no-match states, and review flags;
  • keep all outputs as draft audit evidence for attorney review.

It should not answer legal questions, predict outcomes, decide whether a cause of action exists, draft filed material as final work product, or claim legal compliance. It is a citation-integrity tool that can sit beside legal research connectors, document systems, and practice-area workflows.

A legal MCP connector using Sourceright should stay read-heavy and provenance-first:

RequirementSourceright position
Read-heavy toolslegal.analyze_citations is read-only and operates on a local text file.
Provenance in resultsLegal records include jurisdiction/provider candidates, confidence, and review issues.
No instruction-like retrieved contentProvider evidence is data; it must never be treated as instructions.
Explicit write gatesThe legal citation surface has no write operation. Other MCP writes remain dry-run first and require apply: true.
Graceful failureMissing provider matches and uncertain jurisdictions become review issues, not fabricated certainty.
Attorney reviewEvery output remains draft audit evidence for a reviewing lawyer.
  1. Export or save the legal text to a plain text file.

  2. Run the CLI audit:

    sourceright legal legal-text.txt
  3. Or call the MCP tool from a local stdio client:

    {
    "name": "legal.analyze_citations",
    "arguments": {
    "path": "legal-text.txt"
    }
    }
  4. Review the returned records:

    • citation text;
    • citation type;
    • jurisdiction and court hints;
    • provider candidates;
    • confidence scores;
    • missing-jurisdiction, missing-year, missing-provider, and ambiguous-type issues.
  5. Keep the legal report separate from references.csl.json.

  6. Hand unresolved records to attorney review or to a licensed legal research system. Sourceright should not convert uncertainty into legal conclusions.

The pack is most useful when embedded as a checkpoint inside legal workflows:

WorkflowHow Sourceright helps
Brief section reviewExtract citations and flag missing provider evidence before attorney review.
Demand letter draftingAudit cited authorities before any send gate.
Litigation chronologySeparate case/statute citations from factual source references.
Legal clinic memo reviewShow which cited authorities need verification or jurisdiction review.
Docket or matter intakePreserve citation audit evidence beside the matter file without modifying it.
Claim/source provenanceLink legal citations to cited propositions without asserting claim truth.

CourtListener is the first public legal-provider path for United States case law. The immediate integration model is:

  • fixture-backed CourtListener response parsing by default;
  • optional live lookup only after cache, rate-limit, and auth behavior are documented;
  • explicit matched, no_match, ambiguous, and error states;
  • source URL, retrieved date, and citation-ready identifier in provider evidence;
  • no mutation of academic CSL records.

Other provider lanes should remain separate:

  • AustLII and Australian legislation registers for Australian material;
  • Caselaw Access Project for historical United States cases;
  • official legislation and regulation registers where access terms allow;
  • licensed systems such as Westlaw or Practical Law only as external subscriber-provided research connectors.

A concise connector description for a Claude legal workflow is:

Sourceright audits and enriches legal citations. It extracts legal citation records, identifies jurisdiction/provider candidates, surfaces conflicts and no-match states, and returns provenance-rich review flags. It does not provide legal advice, predict outcomes, or draft final legal work product.

This keeps the value proposition specific enough to be useful to legal workflows while preserving the product philosophy: evidence infrastructure first, attorney judgment last.