Connect an AI client

Excaliself has an MCP server at /api/mcp. Your AI client brings the model, so the instance needs no LLM API key.

1. In Excaliself, open Settings, then MCP, and click New token. Copy the token: you see it only once.
2. Add the server to your client, with your instance’s URL and the token.

Claude Code

The token dialog shows this command filled in:

claude mcp add --transport http excaliself https://draw.example.com/api/mcp \
  --header "Authorization: Bearer exs_..."

Claude Desktop

Claude Desktop cannot send a token to a remote server itself, so it connects through mcp-remote, which needs Node. Open Settings, then Developer, then Edit Config, add this to claude_desktop_config.json, and restart Claude Desktop:

{
  "mcpServers": {
    "excaliself": {
      "command": "npx",
      "args": [
        "-y",
        "mcp-remote",
        "https://draw.example.com/api/mcp",
        "--header",
        "Authorization:${AUTH_HEADER}"
      ],
      "env": { "AUTH_HEADER": "Bearer exs_..." }
    }
  }
}

Other clients

For clients that use Streamable HTTP, the URL is https://<your instance>/api/mcp, with the header Authorization: Bearer <token>.

Tools

• list_drawings, search_drawings, and list_folders find drawings, with their space and folder.
• read_drawing describes a drawing: its shapes, text, and what the arrows connect. It does not send the raw Excalidraw JSON.
• create_drawing makes a new drawing in your private space or in Team, optionally in a folder, from simple shapes and arrows or a Mermaid flowchart.

Access

A token acts as you: it sees what you see. It can create drawings, but it cannot change or delete drawings that exist. Revoke a token under Settings at any time. When an admin deletes a member, that member’s tokens stop working.