MCP tools and project access

Connect servers you use

The plug icon in the editor’s activity bar opens the MCP Servers page. Each server card shows its command and status, with Start and Stop. Selecting a card lists tools, resources and prompts with their descriptions.

Add Server selects an installed program; Add by URL adds a Streamable HTTP endpoint. Configuration is stored in ~/.config/crc/mcp.json and reloaded when you save it in crc. It uses the same mcpServers object shape as Claude Code’s config.

Servers start only when you click Start. Commands that download and run code, such as npx, bunx, uvx and pnpm dlx, are refused. Install and review tools separately before connecting them.

A tool opens a call document. Edit its arguments and press Cmd-Return to run it. A tool marked as able to change things asks first. Save a call to run it again from Home. The server is an external program or service; its access and behaviour are separate from crc’s WASM extension sandbox.

What is supported

crc supports stdio and Streamable HTTP servers. It does not implement OAuth sign-in or answer server requests for sampling, elicitation or roots. Interactive MCP Apps are not rendered; their results appear as text.

Expose a project to agents

crc --mcp [folder] runs a separate stdio MCP server without opening a window. It provides symbols and files from the project index, bounded reads inside the folder, and workspace state and repository information. It is read-only and exposes disk/index state, not unsaved editor buffers.

Run › Copy crc MCP Server Command copies the command to add to an agent’s settings. This server is separate from the Claude Code IDE bridge.