I pointed my routing tester at a server I didn't build. One tool was quietly stealing two others' jobs.

Short summary

Routeproof, a new testing tool, tests how AI models choose between similar MCP tools based on descriptions alone. Testing the canonical MCP filesystem reference server revealed hidden routing bugs that traditional unit tests cannot catch—one tool silently captured requests meant for others. Routing reliability is model-dependent and non-deterministic, requiring multiple test samples and confidence metrics, not green-light passes.

• •routeproof identifies tool-routing bugs that unit tests can't catch
• •Even canonical reference implementations have routing issues when tool descriptions overlap
• •Tool routing is model-dependent and non-deterministic—requires sampling, not single-pass testing

Generated with AI, which can make mistakes.