Taskavel exposes a remote Model Context Protocol (MCP) server for working with the same projects and tasks available in the application. It is useful when you want an AI client to answer questions from live Taskavel data, prepare a plan, or carry out a clearly scoped change without moving between tools.
The production endpoint is:
https://taskavel.com/mcp/taskavel
MCP is an integration surface, not a second Taskavel account. An AI client acts as the Taskavel user who authorizes it or whose personal access token it uses. Project membership and role checks therefore still apply. A client cannot use this connection to see a project that the authenticated user cannot open in Taskavel.
Choose a connection method
Use OAuth when the client presents a browser-based connector flow. Taskavel supports this path for ChatGPT and Claude Desktop. You add the server URL in the client, sign in to Taskavel when asked, and approve the requested connection. The connection can later be removed from Account Settings → AI Integration (MCP).
For command-line and agent configurations, use a personal access token (PAT). Create a named token on the same AI Integration page, store it in the client’s secret or environment-variable configuration, and send it as an Authorization: Bearer credential. PATs are separate from OAuth connections, so revoking one does not revoke the other.
What the server can do
The server includes read tools for projects, task details, deadlines, board snapshots, activity, and assigned work. It also includes write tools for creating and updating tasks, comments, tags, milestones, boards, project members, task relationships, and timers. Some actions have a stronger safety boundary: archiving is reversible, while permanent task deletion applies only to an already archived task and requires explicit confirmation.
Taskavel publishes a generated MCP tools reference from the shipped server definition. It is the source for the available tool list and schemas, rather than a hand-maintained copy. Each tool page explains its inputs, classifies the action as read, write, or destructive, and includes a prompt and a call example. The server also provides the Plan my day and Triage new bugs prompt workflows.
A safe first session
Start with a bounded request such as “List my projects” or “Show my tasks due this week.” Let the client resolve project names and inspect the relevant task before asking it to write. For changes, name the project and desired outcome: “In Website, move task #42 to Review.” If the result is ambiguous, the client should ask instead of guessing.
Read Connect to Taskavel MCP to select a setup, then review permissions and project scope before giving an agent write access.