@toolmark/mcp

Desktop MCP for Toolmark: lets a desktop MCP client (Claude Desktop, an IDE, any MCP SDK client) call the tools of an app page open in the user's browser. The toolmark-mcp CLI serves MCP over stdio and runs a pairing WebSocket server on 127.0.0.1; @toolmark/mcp/client provides mcpPairing(), which pairs the page with a short code and serves its registry as caller mcp (policy applies and consequential tools confirm through the app's inline confirm handler).

pnpm add @toolmark/core @toolmark/mcp

ESM only; Node ≥ 22.12 for the CLI. The client entry runs in the browser and never imports ws or node:*.

{
  "command": "npx",
  "args": ["-y", "@toolmark/mcp", "--allow-origin", "http://localhost:5173"]
}

Usage: toolmark-mcp --allow-origin <origin> [--allow-origin <origin> …] [--port <n>] [--call-timeout <ms>]

The CLI prints a pairing code on stderr; the user enters it in the app:

import { mcpPairing } from '@toolmark/mcp/client'

// Resumes from the session token after a reload (inert without one).
let detach = tm.use(mcpPairing({ port: 17840, onStatus: console.log }))

function pair(code: string) {
  detach()
  detach = tm.use(mcpPairing({ code, port: 17840, onStatus: console.log }))
}

Guide (flags, pairing flow, security model): docs/guides/mcp.md.

Docs: MCP guide · API reference.

License

MIT