Skip to main content
Glama
correctover-audit

correctover-audit

Official

correctover-audit-mcp

Free self-service MCP session-security audit as an MCP server. Ask your agent to audit an MCP endpoint and get a re-verifiable verdict (e.g. unauthenticated-surface) from 3 read-only probes — no handshake forgery, no exploit payloads.

Part of the Session-as-Identity research line: DSHCorrectover/mcp-session-audit (protocol + CI action + evidence).

Install (one line)

claude mcp add correctover-audit -- npx -y github:DSHCorrectover/correctover-audit-mcp

Then ask your agent: "audit https://your-server.example/mcp for session-security issues".

Related MCP server: passmcp-server

What it checks

probe

meaning

initialize (anonymous)

does the server issue a session at all?

tools/list (zero handshake)

can an anonymous client enumerate the full tool inventory?

Verdict classes: unauthenticated-surface (exposed) / authenticated-ok / error-*.

Honest scope

  • Read-only probes only; audit endpoints you own or are authorized to test.

  • The verdict is evidence-grade (timestamps + reproduction command), not a certification.

  • Deeper fail-closed review with signed evidence dossiers: wangguigui@correctover.com

MIT © 2026 Guigui Wang (Correctover) — "We sell the verdict, and you can re-verify it yourself."

Tool Schema Changelog

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

Loading tool schema changelog…

    Related MCP Connectors

    Related MCP Servers

    • A
      license
      A
      quality
      B
      maintenance
      Enables AI assistants to run read-only exposure scans against MCP servers that front databases, surfacing unauthenticated access, role/tool leaks, write-capable tools, and locally stored credentials, then returning plain-language findings.
      3
      Apache 2.0
    • A
      license
      B
      quality
      C
      maintenance
      Enables read-only security assessment of a public MCP endpoint by sending initialize and tools/list requests and returning a 0–100 score covering transport, auth, overly broad tools, injection phrasing, and rug-pull definition changes, plus a 12-probe prompt-injection suite against a bot you own that flags leaked system prompts, tool dumps, secrets, exfiltration channels, and canary obedience.
      2
      MIT