Record and replay FrontMCP calls to catch server regressions
Capture baseline MCP calls as JSONL, replay them against a changed FrontMCP server, and review answer differences and new or resolved errors before accepting the build.
npx skills add agentskillexchange/skills --skill record-and-replay-frontmcp-calls-to-catch-server-regressions
Record a known-good FrontMCP server session, replay its MCP calls against a changed build, and produce a reviewable regression report. Invoke this workflow when a coding agent changes a tool, resource, prompt, or server dependency and needs evidence that existing client behavior still works.
What this skill actually does
Inputs and prerequisites
Provide a FrontMCP TypeScript project, representative client calls, a baseline build, the changed server entry point, and a list of fields that vary between runs. Use Node.js 24 or later and the project’s compatible frontmcp, @frontmcp/sdk, and @frontmcp/testing packages. Keep @frontmcp packages on matching versions.
Record a baseline
Enable the SDK’s CallRecorderPlugin for a controlled baseline session, choosing a local JSONL path. Exercise the tools/call, resources/read, and prompts/get requests that matter for the change. Add application-specific sensitive fields to redactFields, then inspect the recording before sharing it or adding it to a repository. Automatic credential redaction does not cover every kind of sensitive data.
Replay against the changed build
Point the project test configuration at the changed server and run npx frontmcp test --replay fixtures/calls.jsonl. Add --replay-ignore '**.generatedAt,**.requestId' only for fields that are expected to vary. The CLI starts a test server and prints a comparison report; it exits with status 1 when a call differs. For authenticated servers, use replaySession() from @frontmcp/testing in a test spec with the appropriate client token and assert with toMatchRecording().
Reviewable output
Return the baseline fixture path, tested revision, exact replay command, ignore rules, exit status, and the report’s differing answer paths. Triage each outcome as same, changed, new-error, or gone-error. Explain intentional changes before accepting a new baseline, and preserve regressions as failing tests.
Scope and limitations
This skill covers regression checks for recorded FrontMCP protocol calls. Replay invokes the server’s operations again, so use isolated test data and dependencies for calls that change state. Redacted inputs can change replay results, and ignored fields reduce comparison coverage. This workflow does not establish production readiness or replace tests for unrecorded behavior.
Follow the upstream recording and replay reference and its parent frontmcp-testing skill for the detailed API and test setup.