Show HN: Reference MCP – let your AI agents search each other's past sessions
The author built Reference MCP, a tool that allows AI agents to search each other's past sessions for context on past decisions. It establishes sessions for direct access when prompted. The project is now available as a repository.
Background
- **MCP (Model Context Protocol)** is an open standard created by Anthropic that lets AI assistants (like Claude) interact with external tools, data sources, or other AI sessions through a defined interface — think of it as a "USB port" for AI agents to plug into each other or into services.<br>- **Claude Code** and **Codex** are two different AI coding assistants: Claude Code is Anthropic's terminal-based coding agent, while GitHub Codex (or "GitHub Copilot Chat") is OpenAI/Microsoft's coding assistant. They don't share memory or conversation history by default.<br>- This project solves a practical pain point: if you were using both AI tools on the same codebase, each would make decisions (why a function was written a certain way, what trade-offs were discussed) that the other couldn't see. The author built a lightweight MCP server that lets one AI agent search and pull context from the other's past sessions on demand, avoiding repeated explanations or contradictory choices.