MCP project management

TaskView ships an MCP server, so Claude, ChatGPT and other AI assistants can plan, update and report on your projects — with exactly the access your role allows.

Set up the MCP server taskview-mcp on npm

Ask in plain language, the agent calls the tools

Each example below maps to tools that exist in the server today.

"What do I have today?"

get_agenda returns today, upcoming, recently completed and undated tasks across every project in one call.

"Turn these meeting notes into tasks"

create_task with priority, deadline and a parent task for subtasks; toggle_task_assignees and toggle_task_tag to hand them out.

"Log two hours on the checkout bug"

log_time, start_timer and stop_timer — and get_time_report or get_time_summary to see where the week went.

"What is blocking the release?"

list_task_dependencies, add_task_dependency and delete_task_dependency work on the same graph you see in the app.

"Move everything reviewed to Done"

update_task changes status, deadline or list; the kanban tools list, create and rename columns.

"Undo what you just changed"

get_task_history shows every change to a task, and restore_task_from_history puts an earlier version back.

69 tools in total: projects and lists, tasks, subtasks and task history, kanban columns, task dependencies, tags, file attachments, time tracking and reports, collaborators and roles, organizations and members, notifications, agenda.

Three ways to connect

Local, via npx

Each person runs the server on their machine with their own API token. Works with the hosted app and with your own instance.

{
  "mcpServers": {
    "taskview": {
      "command": "npx",
      "args": ["-y", "taskview-mcp"],
      "env": {
        "TASKVIEW_URL": "https://api.taskview.tech",
        "TASKVIEW_TOKEN": "tvk_your_token_here"
      }
    }
  }
}

Shared HTTP server

Run the gimanhead/taskview-ce-mcp container next to your API. Nobody needs Node.js; every request carries the caller's own token, and the server keeps no data.

claude mcp add --transport http taskview https://mcp.your-domain.com/mcp \
  --header "Authorization: Bearer tvk_your_token_here"

OAuth for cloud assistants

claude.ai and Claude Desktop custom connectors and ChatGPT connectors cannot take a pasted token. Paste the MCP URL instead, sign in, and pick the permissions and projects on the TaskView consent screen.

RBAC decides what the agent can touch

Giving an AI assistant access to your tracker should not mean giving it the keys to everything.

47 permissions, set per project

Roles combine permissions for projects, lists, tasks, kanban, the dependency graph, integrations, sprints, time tracking, files, automations and analytics. A role in one project has no effect in another.

Token scope ∩ your role

A tvk_ API token is limited to the projects and permissions you select; an OAuth grant to what you tick on the consent screen. Both are intersected with your RBAC role — the agent can never do more than you can.

Agents follow the same rules for roles

Tools such as create_role, toggle_role_permission and toggle_collaborator_roles go through the same server-side checks as the app. If your role cannot manage people in a project, neither can your agent.

Revoke in one click

Every OAuth app is listed under Connected apps with its permissions, projects and last use. Revoking cuts it off immediately; access tokens expire after one hour anyway.

Full permission reference: roles and permissions and API tokens.

What the MCP server does not do

Sprints and automations are not available through MCP — manage them in the app.
New files cannot be uploaded over MCP. Upload in the app, then the agent can attach, rename or detach them.
OAuth for cloud clients needs OAUTH_DYNAMIC_REGISTRATION=true on your API; it is off by default.

Questions

What is an MCP server for project management?

MCP (Model Context Protocol) is an open standard that lets AI assistants call tools. The TaskView MCP server exposes your projects, tasks, boards and time entries as tools, so an assistant can read and change them on your behalf.

Which AI clients can connect?

Any MCP client. The documentation covers Claude Code, Claude Desktop, claude.ai custom connectors, ChatGPT connectors, Cursor and VS Code — over a local stdio process, a shared HTTP server, or OAuth.

Does it work with the hosted app or only self-hosted?

Both. Point TASKVIEW_URL at https://api.taskview.tech for the hosted app, or at your own API for a self-hosted instance.

Can the agent see projects I did not allow?

No. An API token is scoped to the projects you select, and an OAuth grant to the projects you tick on the consent screen. Both are intersected with your role, so the agent never gets more access than you have.

Is the MCP server free?

Yes. It is part of the free Community Edition, published on npm as taskview-mcp and as the gimanhead/taskview-ce-mcp Docker image.

See also project management for software teams and how TaskView compares with Plane and OpenProject.

Connect your assistant

Create a scoped API token in your account settings, add the server to your client config, and ask what is on your plate today.

MCP setup guide Install TaskView