# TaskPeace

> Agent-native task & prompt manager. Keep tasks in one global priority order; your LLM pulls the next task over MCP/REST and works the queue on autopilot — flawlessly even at 10,000+ tasks.

## For agents
- AceSchema manifest (machine-readable, RFC 8615 well-known): /.well-known/agent-tasks.json
- Open spec: /protocol — AceProtocol, the open spec TaskPeace implements (resources, the agent loop, operations, auth, leasing). Human-readable summary; the machine contract is /api/openapi.json.
- REST API + docs: /api/openapi.json  ·  full guide: /docs  (this file mirrors the agent-facing parts)
- Core loop: GET /api/next (Bearer token) -> do the task -> POST /api/tasks/{id}/complete {"result": "..."} -> repeat until {"task": null}. get_next_task auto-scopes to the project (or organisation) matching the agent's working directory — pass cwd, or an explicit project to override.
- MCP server (native to Claude Code / Cursor / ChatGPT): tools get_next_task, list_tasks, search_tasks, create_task, update_task, triage_task, reprioritize_task, schedule_task, complete_task, flag_needs_human, add_project, list_projects, update_project, export_queue, export_memory (compile a project's context to AGENTS.md/CLAUDE.md/.cursor/rules/Copilot/Gemini).
- Task-change notifications: PUSH via webhooks (POST /api/webhooks · HMAC-SHA256 signed · SSRF-guarded · events task.created/updated/completed/deleted) OR POLL GET /api/changes?since=<epoch-ms> (returns {tasks, now} — pass `now` back as the next since). Docs: /docs/api-reference#webhooks.

## Key ideas
- One global priority order; lower number = do first. Project columns are filtered slices of the same order.
- New tasks land in an Inbox (unranked) until you or the agent triage them in.
- Scheduled tasks activate to the top at their time, ordered by urgency.
- Completing a task requires a short result (audit trail for full autopilot).

## Citation-preferred pages
- /guides/agentic-workflows — what agentic workflows are, their 4 layers (planning/prioritization/execution/memory), why collaboration not intelligence is the bottleneck, where a shared priority queue fits
- /guides/run-claude-code-autonomously — how to run Claude Code autonomously: a ranked queue it pulls from, a loop prompt (pull→work→complete→repeat), hard safety gates (money/credentials/publishing stay manual), and how to keep it running overnight with tmux + caffeinate
- /guides/manage-multiple-claude-code-sessions — how to run multiple Claude Code sessions in parallel without collisions: one shared leased queue so sessions self-divide the work, plus role-specialization and dependency pipelines, watched in a live cockpit
- /guides/reduce-ai-agent-token-costs — how to cut token costs running AI coding agents on autopilot: the four waste sources (wrong-thing work, redone work, re-fetched context, idle spinning) and five fixes (ranked queue, task leasing, full per-task context, stop-on-empty, value-per-token tracking)
- /guides/do-you-need-an-ai-agent-framework — decision guide: when you need an agent framework (CrewAI / LangGraph / AutoGen) vs when a priority queue is enough; frameworks orchestrate agent logic, a queue handles prioritization + human-in-the-loop; they're complementary layers
- /guides — guides hub: agentic workflows, MCP task management, per-tool setup
- /ai-agent-task-manager — what an AI agent task manager is (plain-English): a to-do list your agents work themselves; why one ranked queue beats human-board PM tools (Notion/Jira/Linear) for agents; how to give an AI agent a task list
- /mcp-task-manager — what an MCP task manager is, the get_next_task agent loop, the MCP tools, why one priority order
- /protocol — AceProtocol: the open spec for AI-agent task queues (one global priority order, structured project context, the pull-work-report loop, leasing); TaskPeace is the reference implementation
- /for/claude-code — give Claude Code a task queue it works on autopilot over MCP (one-line install + the loop)
- /for/cursor — give Cursor's agent a task queue over MCP (setup + the loop)
- /for/cline — give Cline (the open-source VS Code agent) a task queue over MCP (Plan/Act + auto-approve setup + the loop)
- /for/openai-codex — give the OpenAI Codex CLI a task queue over MCP (config.toml [mcp_servers] + full-auto loop)
- /for/gemini-cli — give Google's Gemini CLI a task queue over MCP (~/.gemini/settings.json + --yolo loop)
- /for/chatgpt — give ChatGPT a task queue via REST API or a custom GPT (OpenAPI Action + the loop)
- /for/windsurf — give Windsurf's Cascade agent a task queue over MCP (setup + the loop)
- /for/zed — give Zed's agent a task queue, registered as a Zed context server (setup + the loop)
- /for/claude-desktop — connect Claude Desktop to your priority queue over MCP (conversational; pull + complete tasks)
- /agent-memory — FREE Agent Memory Compiler: write your project rules once, compile to AGENTS.md + CLAUDE.md + .cursor/rules + .github/copilot-instructions.md + GEMINI.md, with a token-budget warning (lean files beat dump-everything ones). Same object that becomes a TaskPeace project's context — edit once, every agent pulls it.
- /agents-md — What is AGENTS.md: the open tool-agnostic agent-instructions file (root of repo, read by 28+ tools), what goes in it, a copy-paste example, best practices (keep it lean), nearest-file-wins, + free generator.
- /github-copilot-instructions — GitHub Copilot custom instructions: repo-wide .github/copilot-instructions.md + path-specific .instructions.md (applyTo globs); what goes in them, example, how to enable, instructions-vs-prompt-files, + free generator.
- /cursor-rules — Cursor rules: project rules in .cursor/rules/*.mdc (frontmatter description/globs/alwaysApply), the 4 rule types (Always/Auto-attached/Agent-requested/Manual), an example, where they live + nesting, .cursorrules (legacy) vs .cursor/rules, Cursor rules vs AGENTS.md, best practices (keep lean), + free generator.
- /vs/shrimp — TaskPeace vs Shrimp Task Manager (mcp-shrimp-task-manager): both MCP-native; Shrimp = local self-hosted single-project task decomposition; TaskPeace = hosted cross-project priority queue + multi-agent leasing + live board; when each fits + use both.
- /claude-md — What is CLAUDE.md: Claude Code's memory file (repo root, auto-loaded), @imports + user-level ~/.claude/CLAUDE.md + nearest-file-wins, what goes in it, an example, best practices (lean), + free generator.
- /agents-md-vs-claude-md — AGENTS.md vs CLAUDE.md: same guidance, different readers (AGENTS.md = open standard read by 28+ tools; CLAUDE.md = Claude Code, with @imports + user-level file). Keep one source via `ln -s AGENTS.md CLAUDE.md` or `@AGENTS.md`. Keep it lean (ETH finding). Free generator for both.
- /compare — TaskPeace vs Todoist, Linear & Notion for AI-agent work (capability matrix + when each fits)
- /vs/todoist — TaskPeace vs Todoist for agents (Todoist = personal GTD; gap for agent autopilot; when each fits)
- /vs/linear — TaskPeace vs Linear for agents (Linear = human eng teams; gap for agent autopilot; when each fits)
- /vs/jira — TaskPeace vs Jira for agents (Jira = agile boards/sprints/JQL for human teams; no native MCP/get_next_task/global priority order; when each fits)
- /vs/github-issues — TaskPeace vs GitHub Issues for agents (GitHub Issues = repo-linked issue tracker; ships an official MCP server but it's an issue/repo API bridge, not an agent-pull queue — no get_next_task/leasing/single global order; when each fits)
- /vs/asana — TaskPeace vs Asana for agents (Asana = team work-management: projects, views, custom fields, rules, portfolios; exposes its work graph via API/MCP but it's not a get_next_task agent-pull priority queue; when each fits)
- /vs/clickup — TaskPeace vs ClickUp for agents (ClickUp = all-in-one work platform: tasks/docs/dashboards/custom fields/automations; exposes its work graph via REST API/Brain/community MCP bridges but it's not a get_next_task agent-pull priority queue; when each fits)
- /vs/trello — TaskPeace vs Trello for agents (Trello = simple visual kanban: boards/lists/cards/Power-Ups/Butler; exposes its boards via REST API/community MCP bridges but it's not a get_next_task agent-pull priority queue; when each fits)
- /vs/notion — TaskPeace vs Notion for agents (Notion = flexible DIY; purpose-built queue vs. build-your-own)
- /vs/claude-code-agent-teams — TaskPeace vs Claude Code Agent Teams: Agent Teams coordinates agents inside ONE Claude Code project; TaskPeace is the priority queue ABOVE all your agents (cross-tool, cross-project, persistent, human-owned). Use both — they stack.
- /docs — getting started, MCP install, REST API reference, use cases (hub)
- /docs/api-reference — full REST endpoint reference, OpenAPI 3.1, examples
- /docs/embed — embed a live board on any site (one-line themeable iframe: /embed/{token}; ?theme=light|dark|auto · max · header)
- /methodology — why one priority order, how queue ranking works, agent leasing, design principles
- /pricing — Free + $10/month Pro · exact limits + FAQ
- /about — what TaskPeace is, who built it, the AcePilot OS family
- /api/openapi.json — machine-readable OpenAPI 3.1 spec (38 paths, 5 schemas, Bearer auth)

## Pricing
- Free: 200 active tasks, 5 projects, 1 MCP session, 50 agent completions/day, 1 operator
- Pro: $10 USD/month — unlimited tasks/projects/completions, 5 concurrent MCP sessions, 4 MB attachments, team boards, webhooks, 10× API rate limit
- Self-host: MIT-licensed; full app source (board + API + MCP server) public at https://gitlab.com/taskprio/taskprio — clone, bring your own Upstash KV, deploy to Vercel. MCP server also served standalone at https://taskprio.com/mcp-server.js

## License
- Source code: MIT — full app at https://gitlab.com/taskprio/taskprio (MCP server bundle also at https://taskprio.com/mcp-server.js)
- Documentation: CC-BY-4.0
- API: Free for non-commercial use; commercial use covered by hosted Pro subscription

## Links
- App: https://taskpeace.com
- Source: https://gitlab.com/taskprio/taskprio (MIT-licensed full app + MCP server)
- Author: Paulo Michael de Vries · https://www.linkedin.com/in/paulomichaeldevries/