TaskPrio is an agent-native task manager. Write tasks in one global priority order; your AI agent pulls the next-highest one via MCP, works it, and reports back. This page covers everything you need — install your agent, connect the queue, watch it ship.
Open taskprio.com, sign in with Google, and add your first task to the Inbox column. Drag it into the queue to rank it, or use the sidebar's quick-capture box to add a task that lands at the bottom of your priority list. That's it — you have a ranked priority queue.
Want to try without signing up? The same board works in local-only mode — everything stays in your browser. No account, no agent, no problem.
Sign in, open Account from the cloud icon, and click Install MCP (1 command). Paste the copied one-liner into your terminal. The installer verifies your token, downloads the bundled MCP server to ~/.local/share/promptprio/, and merges a promptprio entry into ~/.claude.json (Claude Code's MCP config, via claude mcp add-json). Restart Claude Code. Done.
PROMPTPRIO_API_TOKEN=pp_xxx bash <(curl -fsSL https://promptprio.com/install.sh)/api/auth/memcp-server.js (1.1 MB · single self-contained Node binary)promptprio entry into ~/.claude.json via claude mcp add-json (preserves your other MCP servers)Cursor / Continue / other MCP clients: see the manual JSON snippet in Account → Copy MCP config. Drop into your client's MCP config file.
Never paste your token into chat — only into your own terminal or your own config file.
On every /api/next call (or MCP get_next_task), the agent receives the single highest-priority unblocked task with its full title, body, attachments, subtasks, and project agent-context. TaskPrio leases the task for 15 minutes so two agents can't race. The agent works it, calls complete_task with the result, and the lease auto-releases. The next call pulls the next-highest. Top to bottom, without you in the loop.
When the agent finishes, the card lights up green for 15 minutes (active lease indicator), then auto-clears. If it gets stuck, it flags needs_human and the card turns yellow — click it, answer the question, send back to the queue.
Paste our maintained autopilot prompt into your agent — it adds hard safety gates (money / credentials / publishing stay manual), an empty-queue stop, and auto-scopes to the project your terminal is in.
Any work where you can write the task once and the agent can grind through. Common fleets: solo founders shipping multiple sites, marketing teams running content engines, data teams batching cleanup, ops teams running periodic checks. The constraint is "your AI can do it given the prompt + the codebase access" — not "the AI must already know how."
TaskPrio is an agent-native task manager. You write tasks in one global priority order; your AI agent calls /api/next, gets the highest-ranked task with full context, works it, and reports back — while you do other things.
Sign in at taskprio.com, click Account → Install MCP. Paste the one-line installer into your terminal. It verifies your token, downloads the bundled MCP server, and merges a promptprio entry into ~/.claude.json. Restart Claude Code.
Free is one operator, unlimited tasks, basic MCP access, and metered agent completions per day. Pro ($10/month) adds unlimited agent completions, file attachments up to 4MB, priority API rate limits, and team boards. See privacy and terms.
Yes — anything that speaks Model Context Protocol (MCP) or REST. Claude Code and Cursor consume the MCP server out of the box. ChatGPT and other LLMs can hit the REST API directly (see /docs/api-reference).
Cloud-mode boards live in Upstash KV (encrypted at rest, region-pinned per account). Local-mode boards stay in your browser's localStorage — they never touch any server. You can switch modes any time without losing data. Privacy policy.