feat: Add MCP Server Integration for LLM-agnostic PR Analysis

[ipanov-ritech]

Summary

Add a production-ready MCP (Model Context Protocol) server integration that mirrors the CLI workflow exactly, enabling PR Agent to work with any MCP-compatible tool (Claude Code, Cursor, Windsurf, Cline, etc.).

Motivation

The MCP server allows PR Agent to be used with any LLM-powered tool that supports the Model Context Protocol, without requiring users to configure AI provider API keys. The calling tool's LLM performs the analysis, making PR Agent truly LLM-agnostic.

Implementation Details

New Files

• src/mcp/server.ts - Main MCP server implementation
• src/mcp/index.ts - Module exports
• src/mcp/README.md - MCP-specific documentation
• server.json - MCP Registry configuration
• smithery.json - Smithery marketplace configuration
• docs/MCP-SERVER.md - Comprehensive documentation

Features

• Two MCP Tools:
  • analyze - Mirrors pr-agent analyze CLI command exactly
  • dashboard - Starts web dashboard for analysis history
• Same Configuration - Uses .pragent.config.json like CLI
• Full Feature Parity - Jira peer review, arch-docs, risk detection, complexity scoring
• Database Storage - Saves to pr-agent.db for dashboard
• Publishing Ready - Configured for npm, MCP Registry, and Smithery

Publishing Configuration

• npm package: @techdebtgpt/pr-agent
• MCP Registry name: io.github.techdebtgpt/pr-agent
• Smithery: Ready for deployment

Testing

Tested with:

• Claude Code MCP integration
• PR analysis on external repositories
• Dashboard functionality

Checklist

• MCP server mirrors CLI workflow
• Documentation updated
• Publishing configuration added
• Security best practices followed
• gitignore updated for db files

Assignee

@ipanov-ritech

Related

This implements MCP server support as discussed in the project roadmap.