Skip to main content
Glama
clearon-source-audit

clearon-source-audit

by kwgoodwin

Clearon Source Audit MCP

MCP server for revision-bound source review, quotation matching, and prepublication evidence-control reports.

It is designed to preserve exactly what was audited. It does not publish articles, decide truth, or replace legal/editorial judgment.

What it does

  • Binds each audit to an exact article revision fingerprint.

  • Records source classifications and claim-to-source mappings.

  • Verifies direct quotations against retrieved source text.

  • Revalidates stored source-text fingerprints before trusting retrieved evidence.

  • Flags procedural-language risks such as stating allegations as holdings.

  • Generates immutable JSON reports for the audited revision.

  • Highlights source-integrity mismatches as explicit publish/update review items for the user.

Related MCP server: traceable-research-mcp

What it does not do

  • It does not determine legal correctness or factual truth.

  • It does not publish or modify external systems.

  • It does not treat retrieval failure as proof that a source is false.

  • It does not convert acknowledged warnings into substantive support.

  • It does not treat acknowledged substantive claim gaps as publication-ready support.

  • It does not treat source-integrity mismatches as repaired just because the user approved a publish/update exception.

Requirements

  • Node.js 20 or newer

  • A local MCP client that can launch a stdio server

Installation

Published package:

npm install -g clearon-source-audit-mcp
clearon-source-audit-mcp

Source checkout:

cd tools/clearon-source-audit-mcp
npm install
npm test
npm run smoke

MCP client setup

Example stdio configuration:

Installed package via npx:

{
  "mcpServers": {
    "clearon-source-audit": {
      "command": "npx",
      "args": ["-y", "clearon-source-audit-mcp"]
    }
  }
}

Source checkout:

{
  "mcpServers": {
    "clearon-source-audit": {
      "command": "node",
      "args": ["/absolute/path/to/clearon-source-audit-mcp/server.mjs"]
    }
  }
}

If you want audit data somewhere specific, set CLEARON_SOURCE_AUDIT_ROOT in the client environment.

Storage

By default, audit data is stored outside the repository in a user data directory:

  • macOS: ~/Library/Application Support/clearon-source-audit-mcp/source-audits

  • Linux: ${XDG_DATA_HOME:-~/.local/share}/clearon-source-audit-mcp/source-audits

  • Windows: %APPDATA%\\clearon-source-audit-mcp\\source-audits

Override that location with CLEARON_SOURCE_AUDIT_ROOT.

Each audit stores:

  • audit.json

  • article.txt

  • reports/*.json

Files are written with owner-only permissions where the host platform supports them.

  1. Create or obtain the exact article text to audit.

  2. Compute its SHA-256 fingerprint.

  3. Call create_source_audit.

  4. Call register_source for each source.

  5. Call record_claim_mapping for checkable factual or legal claims.

  6. Run run_mechanical_checks.

  7. Fix weak claims, missing excerpts, quotation mismatches, and procedural warnings in the article.

  8. If the article text changes, create a new audit for the new revision.

  9. Call generate_source_report.

  10. Use acknowledge_finding only when a human has actually reviewed and accepted a claim or source-retrieval warning.

  11. If the report lists pending_publish_update_review_findings, use approve_source_integrity_finding only when you explicitly approve that exact source-integrity mismatch for publish/update.

  12. Treat ready_for_human_approval as blocked until every claim is substantively supported and every non-integrity warning is either resolved or explicitly acknowledged.

  13. Treat ready_for_publish_or_update as blocked until the report has no pending source-integrity review items.

Tools

  • get_server_health

  • create_source_audit

  • get_source_audit

  • list_source_audits

  • register_source

  • record_claim_mapping

  • replace_claim_mapping

  • run_mechanical_checks

  • generate_source_report

  • approve_source_integrity_finding

  • acknowledge_finding

Network behavior

URL availability checks are opt-in. run_mechanical_checks and generate_source_report default check_urls to false.

When enabled, the server performs bounded HTTP(S) reachability checks and treats failures as warnings only. Before each request (including redirects), it rejects local/private DNS answers and pins the connection to the validated public addresses to narrow DNS-rebinding exposure. This is a convenience check, not a trust decision or a replacement for host-level egress controls.

Source-retrieval warnings use a stable acknowledgment string; transport details are preserved separately in the verification output.

Source-integrity warnings are handled differently. They appear in a dedicated publish/update review list in the report and require explicit user approval before the report becomes ready_for_publish_or_update.

Development

npm test
npm run smoke
npm run syntax
node --check server.mjs

Available Tools

9 tools
acknowledge_findingB

Record human review of an exact warning. Does not convert the underlying source or claim into support.

Inputs: note (required), slug (required), type (required), item_id (required), +2 more.

create_source_auditB

Create an audit bound to the exact SHA-256 fingerprint of an editorial revision.

Inputs: slug (required), title (required), revision (required), article_content (required), +2 more.

generate_source_reportB

Generate an immutable prepublication report for the bound revision, listing every unresolved or acknowledged exception. URL availability checks are opt-in.

Inputs: slug (required), check_urls, timeout_ms.

get_server_healthA

Report source-audit health, storage root, and non-publishing safety boundaries.

Inputs: no named parameters.

get_source_auditB

Read an audit, its exact bound article text, registered sources, claims, and acknowledgments.

Inputs: slug (required).

list_source_auditsA

List source audits and compact workflow counts.

Inputs: no named parameters.

record_claim_mappingA

Record a human-reviewable claim-to-source mapping, exact article excerpt, source excerpt, quotation, and procedural posture. Does not decide truth.

Inputs: slug (required), status (required), claim_text (required), procedural_posture (required), +6 more.

register_sourceC

Register and transparently classify a source; optional retrieved text enables quotation matching.

Inputs: url (required), slug (required), title (required), source_type (required), +6 more.

run_mechanical_checksC

Check the revision fingerprint, quotation matches, and procedural-language warnings. URL availability checks are opt-in.

Inputs: slug (required), check_urls, timeout_ms.

Tool Schema Changelog

Recent tool additions, removals, and schema changes observed during successful MCP inspections.

Loading tool schema changelog…

    TDQS

    B3.4/5.0

    Scored across 9 tools

    Disambiguation4/5

    Most tools have clearly distinct purposes, but run_mechanical_checks and generate_source_report both involve opt-in URL availability checks and exception discovery, so an agent could hesitate between the check step and the report step. All other boundaries, such as registering a source versus creating an audit versus recording a claim mapping, are clear.

    Naming Consistency5/5

    All nine tools consistently follow a snake_case verb_noun pattern: register_source, create_source_audit, list_source_audits, run_mechanical_checks, acknowledge_finding, and so on. There are no mixed conventions or ambiguous generic verbs.

    Tool Count5/5

    Nine tools is well-scoped for a source-audit workflow, covering health checks, source registration, audit lifecycle, claim mapping, mechanical verification, report generation, and finding acknowledgement. Each tool occupies a meaningful workflow slot without redundancy or excessive granularity.

    Completeness4/5

    The core audit lifecycle is well covered: register sources, create audits, map claims, run checks, acknowledge findings, and generate reports. A minor gap is that sources and findings cannot be independently listed or managed outside of an audit, though agents can work around this through get/list audits and report output.

    Maintenance

    ActivityInactive
    ResponsivenessOutdated data

    Related MCP Connectors

    Related MCP Servers

    • A
      license
      A
      quality
      B
      maintenance
      Enables scientific due diligence by grading claims against public literature, clinical trials, and filings, with explicit citations and optional attestation.
      4
      MIT
    • A
      license
      Not graded
      quality
      C
      maintenance
      Enables traceable scholarly literature reviews using free APIs, generating reports where every claim links to evidence IDs.
      MIT
    • A
      license
      A
      quality
      B
      maintenance
      Enables scientists and their AI agents to check quoted passages against pinned open-access PubMed Central full texts, look up retraction status from PMC and Crossref, and run recorded literature searches. Every quote check, search and annotation is written to an append-only, hash-chained log that can be verified, giving a dated record of what was searched and found.
      7
      MIT
    • A
      license
      Not graded
      quality
      B
      maintenance
      Enables users to verify citations and bibliographies and to check whether source documents support factual claims, using local inference, exact evidence provenance, and deterministic verdicts.
      1
      MIT