transcripts-mcp

Search your Cursor, Claude Code, and Codex conversations from any MCP client.

The server reads session transcripts already on your machine. Browse recent sessions, retrieve a conversation, or search across tools with fuzzy, full-text, and optional semantic search.

Setup

Requires Node.js 24 or later. Add this configuration to your MCP client:

{
  "mcpServers": {
    "transcripts": {
      "command": "npx",
      "args": ["-y", "transcripts-mcp"]
    }
  }
}

The npm launcher uses a platform binary with grep and full-text search. If the binary is unavailable, it looks for Bun on PATH.

For semantic search, use bunx --bun transcripts-mcp as the launch command. The first semantic index build downloads a model; embeddings are computed locally.

If search_transcripts reports INDEX_REBUILD_REQUIRED after a server upgrade, run build_index with full: true. This replaces only the local search cache, not source transcripts. Rebuild with semantic: true to recreate embeddings. Give concurrently running incompatible server versions separate TRANSCRIPTS_MCP_INDEX paths.

The server does not modify transcripts. Search results are returned to your MCP client, which handles them according to its own settings.

See the README for client-specific setup, tool reference, and configuration. Development instructions are in CONTRIBUTING.md.

License

MIT