MuhammadUsmanGM/McpProbe

MCPProbe has a strong, timely premise: a zero-install TypeScript CLI that probes MCP servers, detects stdio vs HTTP/SSE, extracts tools, scores server quality, checks client compatibility, and generates copy-paste configs for tools like Claude, Cursor, Windsurf, Codex, and others. The project is small enough to understand quickly, but it already has a sensible internal shape: src/commands/probe.ts orchestrates the workflow, while fetching, transport detection, connection, scoring, compatibility, config generation, and output are split into focused modules. That makes the codebase approachable for contributors and gives the tool a clear path to grow as the MCP ecosystem changes.

The README does a good job selling the practical use case and the npx mcpprobe <target> workflow is easy to grasp. I also liked seeing an MIT license, a changelog, a contribution guide, and tests for URL parsing, endpoint discovery, config generation, and scoring behavior. Those are the right areas to test for a CLI that depends heavily on heuristics and generated output.

The biggest thing I would fix before pushing the project harder is CI. .github/workflows/ci.yml currently runs commands inside packages/cli and packages/web, but this repository appears to be a single-package project at the root, so the workflow likely fails as written. I’d update CI to run npm install, npm run lint, npm test, and npm run build from the repository root. I’d also add a few end-to-end tests with a tiny fake MCP stdio server and a fake HTTP endpoint, because the most important promise of the tool is not just that it parses metadata, but that it can reliably connect, list tools, and report meaningful compatibility results. The scoring model is a useful differentiator, but it would be stronger if the README documented how each score is calculated and where users should treat a score as advisory rather than definitive. Overall, this is a focused developer tool with a real niche; tightening CI, adding integration coverage, and documenting the scoring assumptions would make it much easier for MCP server authors to trust it in their workflow.