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
  bellwether init npx @mcp/your-server
  bellwether check
  bellwether baseline save
  # Later: bellwether check --fail-on-drift

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

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.