Kangram MCP
Open app
← Blog

Connect Cursor to Your Task Tracker via MCP (2026 Guide)

Step-by-step: connect Cursor to Kangram's MCP server in under 2 minutes. Create tasks, move cards, and query your board with natural language.

Kangram 5 min read

Connect Cursor to Your Task Tracker via MCP

Cursor became an agent the moment it learned to speak MCP. One config file, one API key, and your IDE creates tasks, moves cards, and reads your board. No SDK, no boilerplate, no copy-pasting ticket IDs between tabs. This guide connects Cursor to Kangram’s MCP server in under two minutes. The same pattern works for any MCP-compatible task tracker.

Cursor’s MCP support turned the editor from “AI-assisted typing” into “AI-assisted project management.” If you spend your day in Cursor and your tasks in a separate browser tab, you’re paying a context-switch tax on every ticket. MCP removes it. This walkthrough shows how to wire Kangram’s MCP server into Cursor, verify the connection, and run your first natural-language task command.

Prerequisites

You need three things before you start:

  1. Cursor 0.45 or newer — MCP support landed in late 2025 and is stable in current releases.
  2. A Kangram account + API key — sign up at kangram.app, open Settings → Agents, and generate a key with the tasks:write scope (or tasks:read if you only want query access).
  3. Your MCP endpoint URL — also shown on the Agents page. It looks like https://app.kangram.app/mcp.

You do not need Node, Python, or any local runtime. Kangram’s MCP server is remote; Cursor connects to it over HTTPS.

Step 1: Locate your MCP config

Cursor reads MCP servers from ~/.cursor/mcp.json (global) or .cursor/mcp.json inside a project (project-scoped). If neither exists, create one.

mkdir -p ~/.cursor
touch ~/.cursor/mcp.json

Step 2: Add the Kangram server

Open ~/.cursor/mcp.json and paste this block. Replace YOUR_API_KEY with the key from Kangram’s Settings → Agents page.

{
  "mcpServers": {
    "kangram": {
      "url": "https://app.kangram.app/mcp",
      "headers": {
        "Authorization": "Bearer YOUR_API_KEY"
      }
    }
  }
}

Decision note: we use Bearer auth instead of OAuth because agents don’t browse. A long-lived key with a narrow scope (e.g. tasks:write boards:read) is safer in practice than a refresh-token dance that breaks mid-session.

Save the file and restart Cursor (or run the “Reload Window” command). Cursor reads MCP config once at startup.

Step 3: Verify the connection

Open the Cursor agent panel (Cmd+L on macOS, Ctrl+L elsewhere). You should see “kangram” in the connected MCP servers list, with a green status dot and a count of available tools (Kangram exposes ~15: create_task, update_task, list_tasks, move_task, list_boards, and so on).

If the dot is red or the server doesn’t appear, check:

  • The API key is valid and not revoked (Kangram → Settings → Agents shows active keys).
  • The URL has no trailing slash.
  • Cursor fully restarted, not just reloaded the window.

Step 4: Your first natural-language command

With the server connected, type this into the agent panel:

List my in-progress tasks on the DEV board.

Cursor translates that into an MCP list_tasks call with status: "In Progress" and board_id: <DEV board id>. You’ll see the task list inline.

Now try a write:

Create a task "Wire up Stripe webhook idempotency" in the DEV board, priority major, assigned to me.

Cursor calls create_task. The task appears in Kangram instantly — same as if you’d created it from the web UI or the Telegram bot. The audit log records the agent as the actor.

What Cursor can and can’t do via MCP

Can:

  • Create, update, move, and delete tasks
  • Add comments and checklist items
  • Search across boards with filters
  • List boards, statuses, and members
  • Read task activity history

Can’t (by design):

  • Anything outside the scopes on your API key
  • Operations on boards you’re not a member of
  • Mutations on archived boards (the server returns 409)

Decision note: every Kangram command goes through one authorization layer (runUclCommand) regardless of which adapter called it — web, Telegram, REST, or MCP. The agent has no special back door. See our longer writeup on MCP vs REST API for why this matters.

Tips

Scope narrowly. A tasks:read-only key is fine for status-check prompts. Reserve tasks:write for agents you trust to mutate state. You can rotate keys without reconnecting other agents.

Pair with the Telegram bot. Cursor handles work-time prompts. The Kangram Telegram bot handles mobile / async. Both write to the same board. We wrote about the unified-assistant pattern in Building a unified AI assistant.

Check the activity tab. Every agent action lands in the task’s Activity timeline. If something looks wrong, you can trace it back to the exact command.

FAQ

Q: How do I connect Cursor to an MCP server? A: Add the server URL and Bearer token to ~/.cursor/mcp.json under mcpServers, restart Cursor, and the tools appear in the agent panel.

Q: Can Cursor create and update tasks via MCP? A: Yes. A connected Cursor instance can call create_task, update_task, move_task, and every other task command the MCP server exposes, using natural language.

Q: Is MCP support in Cursor free? A: MCP is included with the Cursor editor. The MCP server you connect to (Kangram, GitHub, etc.) may have its own pricing; Kangram’s free tier includes MCP access.

Q: How do I revoke an agent’s access? A: Open Kangram → Settings → Agents and revoke the key. The connection dies on the next MCP call.

Q: Can one Cursor project talk to multiple boards? A: Yes. Specify the board in the prompt (“on the DEV board”) or pass the board ID. You can also create multiple Kangram MCP entries in mcp.json with different keys and scopes.

Conclusion

MCP turned Cursor from an editor into a project management surface. Two minutes of config, one API key, and your agent writes to the same task board your team and your Telegram bot use. No drift, no duplicate state, no context-switch tax.

→ Get your Kangram MCP key · MCP vs REST API: which should your agent use? · Building a unified AI assistant