mcpscore 1.1.0 is here! Our first official release

MCP servers often fail quietly. A stale protocol version, an ambiguous tool description, a broken schema, or incomplete OAuth metadata may not crash the server—it simply makes the experience worse when an agent tries to use it. mcpscore makes those gaps visible before users find them.

Version 1.1.0 evaluates 52 deterministic rules across Protocol, Tools Quality, Security & Auth, and Readiness. Each finding has a stable rule ID, a severity, an explanation, and a link to the MCP specification or RFC behind it. The result is not just a number; it is a prioritized list of what to fix next.

What began as a small command-line auditor in the 0.x era is now a complete quality loop: run it locally while building, enforce it in CI before merging, or check a public server instantly on the web. Reaching 1.1.0 is a huge milestone for us—and the real work starts now!

mcpscore 1.1.0 ships with support for the final MCP 2026-07-28 specification. It recognizes legacy, modern, and dual-era servers, applies version-aware checks, and reports readiness for the new stateless lifecycle without incorrectly judging older servers by rules that do not apply to them.

The release checks the new discovery flow, request metadata, routing headers, result and cache metadata, revised error codes, removed methods, and lingering session behavior. It also corrects capability scoring so optional MCP features are treated as optional while declared capabilities still have to work.

Try the release directly from your terminal:

uvx mcpscore==1.1.0 https://your-server.example/mcp

mcpscore supports local Python and Node.js servers over stdio, plus remote Streamable HTTP and SSE endpoints. Use --json for a machine-readable report, add the GitHub Action to stop regressions in pull requests, or scan a public endpoint on mcpscore.dev. All three surfaces use the same scoring engine.

Audits are designed to be safe: mcpscore inspects advertised capabilities and sends protocol probes, but never invokes a server’s tools.

Our mission is ambitious and simple: help the tech community dramatically raise the quality of MCP servers everywhere. MCP has become a shared foundation for connecting AI to the world at extraordinary speed. That foundation is worth protecting.

It is easy to build an MCP server—and just as easy to build one that is non-compliant, insecure, or difficult for agents to use. Those failures often stay invisible until they happen inside someone else’s agent. We want every team to catch them first, improve with confidence, and keep pace as the standard evolves.

That is why mcpscore is built to be everywhere developers already work: on your machine, in CI/CD, and on the web. Read our full mission and help us make MCP servers better, everywhere!

mcpscore is MIT-licensed and fully open source. The engine, rule implementations, scoring methodology, tests, and release history are all available to inspect. That transparency matters for a tool that tells teams whether their MCP servers are ready to ship.

Explore the complete technical details on the PyPI release page, read the GitHub release, browse the 52-rule reference, or jump directly into the source repository.

This is our first stable release, but it is only the starting line. We’re excited to keep expanding protocol coverage, learning from real servers, and making high-quality MCP implementations easier to build. Let’s raise the bar for MCP together!