Built for the people who ship, connect and sign off on MCP servers.

MCP server authors

Know your server is safe to publish, and show it.

You publish an MCP server, open source or as part of your product. Users and registries are starting to ask whether it is safe, and you would rather find the problems before they do.

What mcpprobe does

• Tests the server before every release with one command, locally or against staging: 16 read-only checks covering authentication, transport, sessions and tool descriptions.
• Runs in CI with the GitHub Action and fails the build at the severity you choose. Results appear in GitHub code scanning.
• Works for remote servers and for stdio servers you run on your own machine.
• Once hosted scanning opens (mid-November 2026, planned), keeps a public README badge that links to the evidence and goes grey if you stop testing.

What you get

A list of findings, each with the request, the response and the fix, before anyone else finds them.

What to run

# a remote server
mcpprobe https://your-server.example/mcp \
  --fail-on high --sarif mcpprobe.sarif

# a local stdio server you run yourself
mcpprobe --stdio "npx -y @your-scope/your-mcp-server@1.2.3"

Public open-source MCP servers get the Team plan free. See the offer.

Teams connecting AI agents

Know which MCP servers your agents can safely use.

Your developers and agents connect to internal and third-party MCP servers. You need to decide which ones to allow, and to notice when one changes after you approved it.

What mcpprobe does

• Tests every MCP server you run before an agent can reach it, as a gate in CI.
• Stores a fingerprint of each server's tool list and flags a tool description that changes after you approved it (META-02).
• Once hosted scanning opens (mid-November 2026, planned), re-tests on a weekly schedule and alerts you when something changes.
• Third-party servers can only be tested with their owner's permission. For those, ask the vendor for their report.

What you get

A CI gate for the servers you own, and a record of what each server exposed when you approved it.

What to run

name: MCP security
on: [push, pull_request]
permissions:
  contents: read
  security-events: write
jobs:
  mcpprobe:
    runs-on: ubuntu-latest
    steps:
      - uses: actions/checkout@v4
      - uses: mcpprobe-dev/mcpprobe@v0.1.0
        with:
          target: https://staging.your-server.example/mcp
          fail-on: high
          sarif-output: mcpprobe.sarif
      - uses: github/codeql-action/upload-sarif@v3
        if: always()
        with:
          sarif_file: mcpprobe.sarif
          category: mcpprobe

Track tool-list changes locally with --baseline .mcpprobe/baseline.json.

Security and compliance

Evidence MCP risk for audits and questionnaires.

Customers and auditors now ask about AI agents and MCP in security questionnaires. You need evidence you can attach to an answer, not a statement of intent.

What mcpprobe does

• Maps every finding to the OWASP MCP Top 10 (beta), the OWASP Top 10 for Agentic Applications and the OWASP Top 10 for LLM Applications. Mapped to, not certified against.
• Records the exact request and response behind every finding, and lists the items a scan cannot see as things to confirm yourself.
• Once hosted scanning opens (mid-November 2026, planned), re-tests daily on the Business plan, with evidence export.

What you get

A JSON report and SARIF file today; scheduled evidence on hosted plans.

What to run

mcpprobe https://your-server.example/mcp \
  --json mcpprobe.json --sarif mcpprobe.sarif

See what a report looks like in the sample report, and how each item maps in the OWASP coverage table.

Not a fit

What mcpprobe is not.