The MCP task manager for AI agents

TaskPrio is an MCP-native task manager: one global priority queue your AI agents read and complete over the Model Context Protocol. Your agent calls get_next_task, does the work, reports a result, and pulls the next one — running the queue on autopilot.

What is an MCP task manager?

An MCP task manager exposes its task queue to AI agents over the Model Context Protocol. Traditional to-do apps were built for humans clicking checkboxes. An MCP task manager is built so an autonomous agent can pull work itself: ask for the next task, receive it with full context, finish it, and report back — with no copy-pasting of prompts between you and the model.

TaskPrio keeps exactly one ranked queue across every project. Lower number = do first. That single order is the answer to "what's next" — for you and for every agent connected to your board.

The agent loop

get_next_task            # highest-priority unblocked task + merged context (leased)
  → do the work
  → complete_task(id, result) # short result = audit trail for full autopilot
  → repeat until the queue returns no task

Because each task is leased when handed out, several agents can work the same queue at once without two of them grabbing the same item.

MCP tools TaskPrio exposes

• get_next_task — pull the top unblocked task with context
• create_task · triage_task · reprioritize_task — capture and rank
• schedule_task — surface a task at the top at a set time
• complete_task — finish with a required result
• flag_needs_human — hand a blocker back to you
• list_tasks · search_tasks · export_queue · add_project

Connect in one line

Claude Code and Cursor install the bundled MCP server with a single command; restart your editor and the tools above are available to your agent.

curl -fsSL https://promptprio.com/install.sh | sh

See the Claude Code and Cursor setup guides, the full docs, or the REST API reference.

Why one global priority order

Notion, Linear, Jira and Trello let you keep many priorities — three "P0" labels, four "Today" columns, two "Urgent" tags. That collapses into "I don't know what to do next" the moment you (or an agent) sit down to work. TaskPrio's single queue removes the ambiguity: there is always exactly one next task. See how it compares →

Frequently asked questions

What is an MCP task manager?

A task manager that exposes its queue to AI agents over the Model Context Protocol, so an agent can call get_next_task, work the top item, report a result, and repeat — running the queue on autopilot. TaskPrio is MCP-native by design.

How does an AI agent get the next task?

It calls get_next_task (or GET /api/next over REST). TaskPrio returns the single highest-ranked unblocked task with merged context and leases it, so no other agent picks it up.

Which tools work with it?

Claude Code, Cursor and ChatGPT — anything that speaks MCP or REST. The two editors use a one-line MCP install; other LLMs hit the REST API directly.

Is it free?

Yes — Free covers 200 tasks, 5 projects and 50 agent completions/day. Pro is $10/month for unlimited use, team boards and webhooks. The source is open (MIT) and self-hostable.

Start free — no card →