null.selectfree assessor

Free MCP risk assessor

Understand an MCP server before you trust it.See the trust gaps before an agent sees the tools.

Enter a public MCP address or paste its tool definitions. You will get clear findings, why each one matters, and what to fix. We read advertised metadata only and never call a tool.

Enter a public HTTPS MCP endpoint or paste a tools/list response. Get deterministic findings for the MCP 2026-07-28 contract, each with a stable code, a reason, a repair, and an exact source location. Endpoint mode reads advertised metadata once and never calls a tool. No account or credential is required.

URLs and definitions are not stored up to 1 MiB · 500 tools same rules as release review

Input

Check a public MCP addressObserve a public MCP endpoint

metadata only · no tool calls

We read the server's name and advertised tools once, then stop. We do not send credentials or call a tool.

We initialize one MCP session, read tools/list, and stop. No credentials, custom headers, resources, prompts, or tool calls are sent.

The URL and discovered definitions are not stored. ⌘ ↵ submits.

What you get

A report you can act on, not a score.

Every finding names what was observed, why it matters, how to repair it, and where in your document it came from. Controls that static metadata cannot prove stay marked not observable.

  1. 01Findings with stable codes, severity, repair, and exact location.
  2. 02Tool-by-tool summary with the declared effect of each tool.
  3. 03Runtime coverage, listing the five controls a document cannot show.
  4. 04Export as JSON for diffing or Markdown for a ticket.
Static inspection is not runtime certification.Endpoint URLs and definitions are not retained.