# mcpprobe

> mcpprobe tests MCP servers for the issues in the OWASP MCP, Agentic and LLM Top 10, records the real request and response behind every finding, and tells you the fix. Open-source CLI, SARIF for code scanning, a badge your users can check.

mcpprobe is a security scanner for MCP servers. Scan only servers you own or are authorised to test. The default passive mode never calls a tool; active checks are gated behind ownership verification and are coming soon. Every finding includes the recorded request and response and a fix. Current CLI version: 0.1.0. Checks: 29.

## Docs

- [Docs (Markdown)](https://mcpprobe.tech/docs.md): quickstart, ownership verification, passive vs active mode, CLI flags, exit codes, config file, GitHub Action, SARIF, badge and status API
- [Docs (HTML)](https://mcpprobe.tech/docs)
- [Checks catalogue](https://mcpprobe.tech/checks): all 29 checks with probe, mode, severity, OWASP mapping and false-positive rate
- [How it works](https://mcpprobe.tech/how-it-works): verify, scan, report, re-scan

## Policies

- [Scanning policy](https://mcpprobe.tech/scanning-policy): what a scan does, never does, how it identifies itself, and how to opt out
- [Security](https://mcpprobe.tech/security): data location, retention, sub-processors
- [Vulnerability disclosure](https://mcpprobe.tech/responsible-disclosure)
- [Terms](https://mcpprobe.tech/terms), [Privacy](https://mcpprobe.tech/privacy), [Acceptable use](https://mcpprobe.tech/acceptable-use)

## Machine-readable endpoints

- [Scanner IP ranges](https://mcpprobe.tech/scanning-ips.json)
- [Status API example](https://mcpprobe.tech/api/v1/status/mcp.demo-company.example): GET /api/v1/status/{host}
- [Badge example](https://mcpprobe.tech/badge/mcp.demo-company.example.svg): GET /badge/{host}.svg
- [security.txt](https://mcpprobe.tech/.well-known/security.txt)

## Blog

- [MCP security checklist for 2026](https://mcpprobe.tech/blog/mcp-security-checklist-2026): A checklist for MCP server authors, mapped to OWASP MCP Top 10 and to the checks that verify each item: authentication, transport, tool metadata, tool code, supply chain and governance.
- [Tool-description poisoning, explained with the actual payloads](https://mcpprobe.tech/blog/tool-description-poisoning-payloads): What a poisoned MCP tool description looks like, why the tool never has to be called, how rug pulls and cross-server shadowing work, and how to detect them without drowning in false positives.
- [How to add OAuth to an MCP server the way the spec expects](https://mcpprobe.tech/blog/mcp-server-oauth-the-spec-way): Protected-resource metadata, the WWW-Authenticate challenge, PKCE, audience validation and no token passthrough: the MCP authorization spec, step by step, with code and tests.

## Optional

- [Pricing](https://mcpprobe.tech/pricing)
- [Changelog](https://mcpprobe.tech/changelog)
- [Compared with mcp-scan](https://mcpprobe.tech/compare/mcp-scan)
- [Compared with Cisco mcp-scanner](https://mcpprobe.tech/compare/cisco-mcp-scanner)