Definition: Taskade as MCP Server is the side of the Model Context Protocol implementation where Taskade exposes its workspace — projects, tasks, documents, agents, automations — as a set of MCP tools that any external AI client can call. Connect Claude Desktop, Cursor, VS Code, or any custom MCP-capable agent to https://www.taskade.com/mcp, sign in once via OAuth2 + PKCE, and the external agent can now list your spaces, read your projects, agents, and automations, create new projects and agents, and edit your Taskade Genesis app source, all through a single standardized interface.
This is the inbound half of the protocol. The mirror direction — an MCP client that calls out to tools hosted on someone else's server — runs through your automations, where the MCP Client step lists and calls a remote server's tools on every plan. An AI agent cannot reach a remote server on its own, so it hands that call to a step. For services that do not speak MCP at all, the 100+ integrations and the built-in agent tools cover the same ground.
Why This Exists
Every AI tool wants to be your primary interface. Most fail because they cannot see the work you actually do. The Model Context Protocol — Anthropic's open standard, donated to the Linux Foundation in late 2025 — fixes this by giving every AI client a uniform way to reach every tool and workspace.
For Taskade users, MCP Server means your Taskade workspace is no longer trapped behind Taskade's UI. Open Claude Desktop, ask "What's on my Q2 roadmap in Taskade?", and Claude reaches into your actual workspace to answer. Open Cursor, point it at a Taskade Genesis app, and let it read the project and edit the app's source files. The workspace travels with you to whichever AI client you prefer.
Architecture
The external client speaks MCP over HTTP with Server-Sent Events. Authentication uses OAuth2 with PKCE — the modern, browser-based, proof-of-authenticity pattern Taskade uses across its web app. Once signed in, the client sees Taskade's tool catalog and can invoke tools over the same connection.
The Tool Catalog
A connected client can act on your workspace, not only read it. What that looks like in practice:
| You ask for | What happens |
|---|---|
| "What's in my Q2 roadmap?" | Your assistant reads the project: tasks, owners, due dates, tags |
| "Turn these notes into a project" | A structured project appears in your workspace |
| "Which agents do I have, and what did that one conclude?" | It lists your AI agents and reads their past chats |
| "Build me an agent that screens service requests" | The agent is created, ready to use |
| "Give that agent our handbook as background" | A project becomes the agent's knowledge source |
| "Ask my analyst agent for the top three risks" | You get the agent's answer without leaving your editor |
| "What's running on a schedule here?" | It lists your automations |
| "Add a status filter to the dashboard" | It edits your Taskade Genesis app |
The catalog keeps growing. It stood at 18 tools in July 2026, up from 3 at launch, and more land with each release.
The write surface is shaped, not unlimited. Your assistant can create projects and agents and edit app code, but it cannot rewrite the contents of an existing project. To change project data, it creates a new one. App edits flow through the same pipeline Taskade EVE uses internally, so Taskade's own agents react to them exactly as they would to a change made in the app. Consistency is automatic.
Everything respects your permissions: an assistant sees only what your Taskade account can see, and editing app code requires Editor access or above.
What This Enables
Claude Desktop as a Taskade cockpit. Pin Taskade's MCP connection in Claude Desktop and use Claude's reasoning across your full workspace without switching apps.
Cursor editing Taskade Genesis app source. Point Cursor at a Taskade Genesis app, let it read the app, propose changes, and apply them — the same developer loop as editing a local codebase.
VS Code as a hybrid IDE. Run VS Code against both your code and your Taskade docs. One agent sees both.
Custom agents for enterprise. Build a custom MCP client that queries both your internal databases and your Taskade projects in a single agent turn. Compliance-friendly, audit-logged, and federated.
Plan Access
The hosted MCP Server is in Beta and available on every paid plan. The Free plan does not include it. If you do not see MCP in your Taskade settings, check pricing for the current paid tiers.
Security Considerations
Taskade's MCP Server uses:
- OAuth2 + PKCE — the external client never sees your password, and connects with a Bearer token carrying the
mcpscope (Personal Access Tokens are rejected on the hosted server) - Per-session tokens — revocable from your Taskade settings
- Scope enforcement — the client only reaches the spaces, projects, agents, and automations your account can already see, and it acts within your role: creating a project or an agent is allowed, editing app source needs Editor access or above, and rewriting an existing project's contents is not exposed at all
- Audit log — every MCP-initiated source edit lands in your workspace activity feed
If you want to revoke access, open your Taskade settings and disconnect the MCP session. The external client immediately loses its connection.
MCP Server vs MCP Client
MCP defines two roles, and every connection pairs one of each. A server publishes a tool catalog and waits to be called. A client discovers those tools and invokes them mid-turn.
On this page, Taskade is the server. Claude Desktop, Cursor, and VS Code are the clients: they sign in over OAuth2 + PKCE, read Taskade's catalog, and act on your workspace. Direction of travel is external → Taskade, the external client holds the tokens, and the tools on offer are Taskade's. That side is live on every paid plan.
The client role runs the other way, and Taskade fills it from your automations: the MCP Client step registers a third-party server, lists the tools it publishes, and calls the one you pick. It works on every plan, including Free, with your automation run allowance as the only ceiling. The boundary to know is that an AI agent cannot reach a remote server by itself — it makes the judgment call and hands the tool call to a step. For everything that does not speak MCP, use the 100+ integrations or the built-in tools every agent already carries.
Related Concepts
- Model Context Protocol — The underlying standard
- The MCP client role — what it means, and how Taskade fills it
- Tool Use — What MCP enables at the behavior level
- Function Calling — The wire format underneath
- Agentic AI — Why MCP matters
- Taskade EVE — Works on the same app files the MCP tools expose
- Taskade Genesis Apps — What gets built on top
Frequently Asked Questions About Taskade as MCP Server
What is Taskade as MCP Server?
Taskade as MCP Server exposes your Taskade workspace — projects, tasks, documents, agents — as tools that any Model Context Protocol-compatible AI client (Claude Desktop, Cursor, VS Code) can call. Sign in once, and external agents can read your workspace, create projects, build and question agents, and edit your Taskade Genesis app.
How do I connect Claude Desktop to Taskade?
Open Claude Desktop, go to integrations or MCP settings, add Taskade's MCP endpoint (https://www.taskade.com/mcp), and complete the OAuth2 + PKCE sign-in flow. Claude Desktop will then list Taskade tools in its tool catalog and can call them during conversations.
What tools does Taskade MCP expose?
Enough to work, not only look. A connected assistant can read your projects, tasks, agents, and automations; create a project from notes; build, update, and question an AI agent; attach a project as an agent's knowledge; and edit your Taskade Genesis app. It cannot rewrite an existing project's contents. The catalog stood at 18 tools in July 2026 and keeps growing. For teams who want to script against Taskade directly, the separate open-source @taskade/mcp-server package runs locally and covers more of the API surface.
Is MCP available on every Taskade plan?
MCP Server is available on every paid plan. The Free plan does not include it. Check your plan on pricing for current eligibility.
How secure is Taskade MCP Server?
Authentication uses OAuth2 with PKCE, so no password is ever shared. The client connects with a revocable per-session Bearer token scoped to mcp, and Personal Access Tokens are rejected. Scope enforcement keeps it inside what your account can already see, and inside your role: it can create projects and agents and edit your Taskade Genesis app source, but it cannot rewrite an existing project's contents. Every MCP-initiated source edit lands in your audit log.
Further Reading
- Model Context Protocol
- The MCP client role — what it means, and how Taskade fills it
- Taskade EVE — The Taskade Genesis Meta-Agent
- AI Agents v2 Capabilities