{"id":46846797,"title":"Show HN: Bellwether – MCP Server Testing and Drift Detection for CI/CD","points":2,"user":"dotsetgreg","time":1769959428,"time_ago":"7 hours ago","type":"link","content":"

What it does:

bellwether check snapshots your MCP server's schema and compares it against a baseline. If anything changed — tools added/removed, parameters modified, types changed — it fails your CI build before you deploy. No LLM calls, no API costs, runs in seconds.

Key features:

• Zero-cost for CI: The check command is completely free and deterministic. No API keys, no LLM calls.

• Granular exit codes: Returns 0-5 depending on severity (no drift → info → warning → breaking changes → errors → low confidence). Configure your pipeline to fail only on breaking changes, or be stricter.

• Documentation as a side effect: Generates CONTRACT.md and AGENTS.md documenting your server's tools, parameters, and behavior.

• Optional deep testing: bellwether explore uses LLMs to test from multiple personas (security tester, QA engineer, novice user) and catches edge cases that structural checks miss.

  npm install -g @dotsetlabs/bellwether\n  bellwether init npx @mcp/your-server\n  bellwether check\n  bellwether baseline save\n  # Later: bellwether check --fail-on-drift\n

\nGitHub Action:

  - uses: actions/checkout@v4\n  - run: npx @dotsetlabs/bellwether check --fail-on-drift\n

\nWhy this matters:

If you're building AI agents that depend on MCP servers, you need to know when those dependencies change. This is like having API contract tests, but for AI tool schemas.

Docs: https://docs.bellwether.sh

NPM: @dotsetlabs/bellwether

GitHub: https://github.com/dotsetlabs/bellwether

Questions welcome. Curious if others are hitting similar MCP drift issues or if there's a better approach I'm missing.","url":"https://github.com/dotsetlabs/bellwether","domain":"github.com","comments":[],"comments_count":0}