ActMCP
Let your AI agent run, debug, and fix GitHub Actions locally — zero CI round-trips
npx -y github:Isco0819/act-mcpEver run into this?
Like `act` which has 42k GitHub stars and proven demand for running GitHub Actions locally, there is still no MCP interface for AI coding agents. Developers manually copy-paste workflow error logs into Claude or Cursor, wait for suggestions, then re-run `act` themselves — a slow, broken loop that burns context and time.
The fix
An MCP server wrapping `act` that exposes run_job, get_logs, and list_workflows as agent-callable tools, so Claude Code or Cursor can autonomously trigger CI jobs locally, read failures, patch the YAML, and re-run — all inside one agent session without human relay.
What it does
- run_job tool: shell out to `act -j <job>` with configurable env and secrets, return exit code and stdout
- get_logs tool: return structured last-run output so the agent can parse errors and propose fixes
- list_workflows tool: glob .github/workflows/*.yml and return job names so the agent knows what it can trigger