I Built a Verification Engine for MCP Servers
The Problem
The MCP (Model Context Protocol) ecosystem has 1000+ servers on GitHub. But there is a massive trust gap: how do you know which ones actually work?
What I Built
MCP Workbench — a web-based testing platform with a verification engine that gives servers Verified / Failed badges.
Key Features
- Connect MCP servers in your browser (no terminal)
- Test tools visually with JSON console
- Automatic verification: startup, discovery, tool calls, performance
- Public verification badges for READMEs
The Verification Engine
| Test | What It Checks |
|---|---|
| Startup | Clean boot, no errors |
| Discovery | tools/list returns within timeout |
| Tool Calls | Each tool responds correctly |
| Performance | Latency p50/p95/p99 |
| Security | File system scope audit |
Results So Far
Out of 20 servers tested:
- 10 passed all tests
- 7 had tool errors
- 3 failed to start
Try It
- Live demo: https://beads-elsewhere-fighters-saskatchewan.mcp-workbench.uk
- GitHub: https://github.com/s63jp/mcp-workbench
- Beta signup: https://beads-elsewhere-fighters-saskatchewan.mcp-workbench.uk/beta.html
Tech Stack
- Next.js 16 + React 19 + Tailwind CSS
- Python asyncio + WebSocket stdio transport
- Zero budget, fully autonomous development
Built this because I was tired of debugging MCP servers in the terminal. Hope it helps someone else too.
Top comments (1)
Verification badges will only stay meaningful if the tests are specific enough to catch real failure modes. Startup and discovery are a start, but tool calls, permissions, latency, and filesystem boundaries are where trust gets interesting. The badge should probably link to the evidence, not just the verdict.