Summary

Build a custom CI/CD Pipeline MCP server that lets agents trigger builds, read structured pass/fail results, and check deployment status across multiple providers (GitHub Actions, Azure DevOps, Vercel).

Context

Per the Final Architecture (Section 3), this is the highest-priority custom MCP. It closes the agent feedback loop: after the Lead Dev writes code and QA approves, the agent can trigger CI, wait for results, and react to failures — all without human intervention.

Acceptance Criteria

• Custom MCP server with tools: trigger_build, get_build_status, get_build_logs, list_recent_builds
• Multi-provider support: GitHub Actions (primary), Azure DevOps, Vercel
• Structured results: pass/fail, failed test names, error summaries (not raw logs)
• Registered in mcp-config.json
• MCPToolProvider spawns and communicates with it
• Integration tests with GitHub Actions API

Depends On

• Issue feat: MCPToolProvider — swap LocalToolProvider for real MCP server subprocess spawning #13 (MCPToolProvider) — must be working first
• Issue feat: GitHub Actions CI pipeline #22 (GitHub Actions CI) — CI must exist for the MCP to trigger

Effort

Large (3+ sessions)