# MCP server

> Let Claude, Cursor and other AI clients find, read, create and update your Taskadence tasks — within the scopes you approve.

**What this is:** a Model Context Protocol server that lets an AI assistant or agent work in Taskadence for you. **When you need it:** you want Claude, Cursor or another MCP client to look up, create or update tasks without leaving the conversation.

> **Available at launch.** The hosted MCP server and the `npx`/`uvx` packages go live with the Taskadence launch. Until then they aren't reachable.

**Claude Code**


```bash
claude mcp add --transport http taskadence "https://api.taskadence.com/mcp"
```

**Any MCP client**


**Server URL**

```text
https://api.taskadence.com/mcp
```

The MCP server speaks to Taskadence only through the API

## What you can do

-   **Find and read work** — search tasks, open one, list yours or a project’s.
-   **Create and change it** — new tasks, status and assignee changes, comments, filing a task under a project.
-   **Stay in control** — the client acts within the scopes you approve, and a task an agent creates waits for a person to review it.

That is 52 tools in 12 groups — see the [tools reference](https://docs.taskadence.com/mcp/tools/).

## Remote or local

The **remote server** is one URL that every client shares, over Streamable HTTP: you add it, sign in to Taskadence, choose your organization and approve the scopes — nothing to install. Clients that can only start a local process use a **stdio package** instead, `npx -y @taskadence/mcp` or `uvx taskadence-mcp`, which proxies the remote server — so the same tools — and signs in with an access token. Either way the server works through the [Taskadence API](https://docs.taskadence.com/reference/), so the same scopes, rate limits and audit trail apply as to any other integration.

---
Source: https://docs.taskadence.com/mcp/