Skip to content

Unknownis not pass

evidence-report-mcp gives agents an evidence report before they act: what was checked, where it came from, and what was not checked.

Get started

Ask before acting

  1. First, add it to Claude Code.

    claude mcp add evidence-report -- npx -y evidence-report-mcp
    toolverify: Verify before acting
    moderead-only
  2. Then let the agent hand over what it is about to use.

    { "kind": "package", "ecosystem": "npm",
      "name": "lodash", "version": "4.17.20" }
    verdictfail Failed: 1/1 checks completed, 4 finding(s).
    sourceOSV.dev, retrieved 19 Nov 2026 11:17:18 UTC
    highCommand Injection in lodash GHSA-35jh-r3h4-6jhm
    highCode Injection via _.template imports key names GHSA-r5fr-rjxr-66jc
    mediumRegular Expression Denial of Service (ReDoS) GHSA-29mw-wpgm-hmr9
    mediumPrototype Pollution in _.unset, _.omit and _.pullAt GHSA-xxjr-mmjv-4gpg
  3. Read where the answer stops.

    not checked
    • Unpublished or zero-day vulnerabilities
    • Malicious behaviour without an advisory (typosquats, install scripts)
    • Whether this package is the right one for the job

Four kinds of target

Packages

Known advisories from OSV.dev.Severity from each advisory's own rating or CVSS vector.

Datasets

Records against their contract, units included.JSON Schema in strict mode, with x-unit: lbf*s is not N*s.

Media files

Content credentials, through c2patool.Only a trusted signer passes. No manifest proves nothing.

Claims

Two independent source domains, at least.Heuristic, so it can fail a claim but never pass one.

One line to add it

claude mcp add evidence-report -- npx -y evidence-report-mcp
Read the guide