Workflow upkeep and run forensics for agentic workflows.
When an agentic workflow fails, the agent can pull its logs, audit the run and compare it with a good one. Before you commit, it compiles edited workflows and applies known fixes.
What it is
The MCP server built into gh-aw, GitHub's command-line extension for agentic workflows. Those are repository automations written in Markdown that run an AI agent inside GitHub Actions, for jobs like issue triage or investigating CI failures. This server lets an agent manage those workflows; it does not run them itself.
What you get
- Each workflow's status, and whether its compiled file is current
- Markdown workflows compiled into GitHub Actions files, with optional security and lint scans
- Workflow run logs downloaded and analyzed
- Runs audited, or several runs compared against the first
- A pull request's CI checks reduced to one clear verdict
- Workflows added from other repositories and updated, with automatic fixes applied
- The MCP servers a workflow uses, and their tools, inspected
Requirements
The GitHub CLI with the gh-aw extension, signed in to GitHub, or GitHub's Docker image with your GitHub token. Running gh aw init sets it up for Copilot and VS Code. Some scans pull Docker images the first time; if you see a download message, wait up to half a minute and retry. Releases from 0.83.3 up to 0.85.4 were retired after a security advisory, so update if you are on one.
Setup effort
One command plus a key — gh aw mcp-server, then supply credentials