What it does
RegressGuard records a baseline of your test suite results, the HTTP status codes and a normalized schema hash for each configured route. Running check reruns the same tests and routes and reports CRITICAL findings (a newly failing test, a changed status code, or a changed response schema) or WARNING findings (response time up more than 200ms and more than 50% over baseline). Schema comparison ignores common dynamic keys such as id, token and timestamp, detects ISO-8601 dates, UUIDs and JWTs, and respects custom ignoreFields.
It is useful when an AI coding agent edits an application and may break an API contract without reporting it. Because it runs as an MCP server, an agent such as Claude Code or Cursor can call check during its own edit loop and fix regressions before the change reaches a developer. The same engine works as a CLI: a git pre-commit hook blocks commits with critical regressions, and a non-zero exit code on critical results works in CI.
Initialization detects the test command, framework and dev server URL. The dev server must be running when the baseline is recorded with snapshot. By default agents get only the check and status tools; the baseline stays under human control. Configuration lives in .regressguard/config.json and supports bearer, cookie or public-route authentication. Every tool call is written to an append-only audit log.
Limitations and requirements
The dev server must be running when the baseline is recorded. Authentication is limited to bearer and cookie modes, or public routes only. Agents receive no snapshot tool by default; enabling agent re-baselining is a configuration change that the source describes as letting an agent accept its own regressions. Bypassing the commit check with git commit --no-verify is possible and the publisher advises it only when accepting the risk.
Safety
Our team has not reviewed this resource. Only automated rules checked it; resources with high-risk signals are not published. These are signals, not a guarantee that installing it is safe. Try it in a sandbox project first.
- Runs code or changes an assistant's behaviour
Sources
- Official MCP Registry first seen 13 Nov 2026 · last seen 13 Nov 2026 · its page on the source
io.github.Bharath-code/regressguard
Deterministic API regression checks for AI agents: snapshot a baseline, catch broken contracts.
v0.2.4As shown on registry.modelcontextprotocol.io when we last read it