Integrations & MCP
Integrations & MCP
Connect Poly to AI assistants, code editors, and other tools using the Model Context Protocol.
Poly integrates with the AI tools and editors you already use. The primary mechanism is the Model Context Protocol (MCP) — a standard that lets any MCP-compatible AI assistant read and interact with your Poly files directly: searching your storage, reading file contents, and organizing files on your behalf. No copy-pasting, no manual uploads.
Two ways to connect
- The polymcp plugin — github.com/withpoly/polymcp packages Poly for specific tools (Claude Code, Claude's desktop app, Gemini CLI) and includes the
poly-cliagent skill. Where the Poly desktop app is installed, these integrations drive the localpolyCLI — fast, offline-capable, and able to use sync-folder paths. - The hosted MCP server — any other MCP client connects to
https://mcp.poly.app/mcp(streamable HTTP). Authentication is OAuth: a browser window signs you into Poly on first connection. There are no API keys to generate or manage.
Available integrations
- Claude Code — the polymcp plugin for your coding agent
- Claude CoWork (Desktop) — one-click extension installed from the Poly desktop app
- ChatGPT Work (Desktop) — the hosted MCP server as a connector
- ChatGPT — the same connector, on the web
- Gemini — Gemini CLI extension and other Google AI tools
- VS Code & Cursor — MCP config for coding with your files
Then read Using MCP — what the search, fetch, and cli tools do, and how to troubleshoot an agent that's misbehaving.
Safety
Integrations can never permanently delete your files: destructive commands are stripped of their force option, so removals archive instead of delete, and anything an agent removes is recoverable — see Recovering Files.