Taskade speaks MCP (Model Context Protocol) in both directions. Pick by what you want to do — the transport and auth details are secondary.
Inbound: an AI client reaching into Taskade
| I want to… | Use | Where / auth |
|---|---|---|
| Have Claude read & write my tasks — create, complete, assign, set dates and fields | Workspace MCP (@taskade/mcp-server) |
Local stdio · personal token (settings/api) · any paid plan |
| Orchestrate my workspace from my IDE — create projects, manage & prompt agents, edit my app's code | Hosted Taskade MCP | Hosted · OAuth 2.0 (https://www.taskade.com/mcp) · any paid plan |
Heads up: The hosted Taskade MCP now creates projects, manages and prompts agents, and edits your Taskade Genesis app's code — but it has no task-level write tools (create/complete/assign/date a task) and no media upload. For task-level writes, use Workspace MCP.
Outbound: Taskade calling another MCP server
| I want to… | Use | Where / auth |
|---|---|---|
| Call another MCP server's tools from a flow — list its tools, then call one as an automation step | MCP Client connector | In-product automation step · the remote server's own URL and auth |
The outbound direction runs through automations, not through the agent runtime: an agent cannot reach an MCP server on its own, so let it hand the work to an automation that has the MCP Client connector as a step.
For services that don't speak MCP at all — Slack, Gmail, Shopify and the rest — Taskade's 100+ native integrations are the shorter path.