AI Agents & MCP Integration
Rerunner ships an out-of-the-box Model Context Protocol (MCP) stdio server and a 1-command setup wizard so coding agents in Claude Code, Cursor, and Codex can inspect, diff, scrub, and debug failing cassettes autonomously.
1-Command Setup
rerunner setup --clients claude-project,cursor,codex
Before modifying any existing config file (.mcp.json, ~/.claude.json, .cursor/mcp.json, .codex/config.toml), rerunner setup creates a .bak backup and merges the rerunner MCP entry idempotently.
Available MCP Tools (rerunner mcp)
| Tool | Purpose |
|---|---|
inspect_cassette |
Summarize entry count, entry types, tokens, and SHA-256 chain status |
read_cassette_step |
Read a single recorded step (request, response, metadata) |
diff_cassettes |
Compare two cassettes step-by-step to pinpoint prompt/output drift |
fork_and_replay_step |
Truncate a cassette after step N to replay up to N and re-run N+1 live |
scaffold_pytest_regression |
Generate a deterministic pytest regression test from a cassette |
verify_cassette |
Verify SHA-256 hash chain and optional HMAC-SHA256 signature |
scrub_cassette |
Redact PII/secrets and optionally re-chain integrity hashes |
estimate_savings |
Compute model-aware token and USD cost savings |
Plugin Manifests & Skills
Rerunner includes native manifests for agent marketplaces:
server.json(MCP Registry).claude-plugin/marketplace.json&.claude-plugin/plugin.json.codex-plugin/plugin.json&.codex-plugin/mcp.jsonskills/rerunner/SKILL.md