Show HN: Share sessions between Codex and Claude Code (and more) (github.com)

A new open-source MCP server, ai-sessions-mcp, exposes local CLI coding sessions from Claude Code, OpenAI Codex, Gemini CLI and opencode to any MCP‑compatible client so your AI assistants can list, search, read and resume past coding conversations. It’s useful for finding past solutions, resuming interrupted work, reviewing recent activity, and enabling agents to learn from your history without sending data to external services — sessions are read from local files (e.g., ~/.claude/projects/.../*.jsonl, ~/.gemini/tmp/.../session-*.json, ~/.codex/sessions/ and archived_sessions/, and ~/.local/share/opencode/storage/). Technically the server is written mostly with Claude Code, distributed under MIT, and built with Go 1.25+. It supports three core APIs: list (filters: source, project_path, limit), search (BM25 ranking; returns relevance score and ~300‑char contextual snippets; filters: query, source, project_path, limit), and get (paginated session content by session_id/source with page and page_size). Install by building the binary (go build) and pointing your MCP config / Claude Desktop mcpServers entry at the executable. For AI/ML practitioners this creates a practical, privacy-preserving bridge between local conversational coding agents and higher-level automation or agent workflows, enabling context-aware retrieval and continuity across tools.