MCP Pre-Flight

MCP Pre-Flight is a submission audit tool for MCP servers being prepared for Claude and OpenAI directories. It is meant to catch the kinds of mismatches that reviewers reject before a submission goes in, and it presents that check as a three-layer process covering wire-level checks, per-tool invocation, and source-side AST analysis.

Its wire-level tests cover OAuth 2.1 with PKCE, dynamic client registration, authorize, token exchange, refresh, and MCP protocol endpoints including initialize, tools/list, and tools/call, along with malformed-body and unknown-method resilience. At the tool level, every safe-to-invoke tool is actually called with synthesized arguments, and idempotent tools are probed by calling them twice and comparing the results. The source-side analysis is optional and accepts a GitHub repo URL; it runs an AST check to compare handlers against annotations, checks that the README mentions every declared tool, and analyzes response language for vocabulary that contradicts annotations. Findings are tagged by directory, with each issue marked for Claude or OpenAI.

The service is aimed at developers submitting MCP servers to those directories. Its FAQ says the input should be the HTTPS endpoint of a deployed MCP server using Streamable HTTP transport, and that the server should respond to JSON-RPC tools/list. It also notes that if OAuth is present, the full auth flow is tested, and that an optional test bearer token unlocks authenticated MCP checks. Source probing uses the GitHub API, and the page says pre-flight only probes the live endpoint; if source is checked later through the M8ven Trust Index, the repository is cloned in an isolated sandbox and then deleted.

MCP Pre-Flight is offered as a free web tool. The page also states that Claude and OpenAI directories require OAuth for listed MCPs, and it mentions an OpenAI domain challenge file at /.well-known/openai-apps-challenge for OpenAI submissions.