Paca
Self-hosted, AI-native project management where AI agents collaborate as equal teammates in Scrum, replacing Jira, Trello, ClickUp, and Monday.
- Source repo
- Paca-AI/paca
- Stars
- ★ 2k
- Last updated
- 1d ago
- License
- Apache-2.0
- Primary language
- Go
- FA score
- 53/100 · Major gaps
At a glance
- How it runs
- Works with
- Universal · cross-platformCodex · Claude Code
- Cost
- Free, no paid service needed
- Setup effort
- Medium · a few setup steps
- You'll need
- Typical use
- A Scrum Master wants to integrate AI coding agents (e.g., Claude Code) as formal team members participating in sprint planning and task execution.
- Not a fit if
- Teams that don't want to run and upgrade their own server
- Teams that need a vendor-managed cloud service out of the box
- Non-technical teams unable to handle Docker and security themselves
- Source review
- 53/100 · Major gaps
What does this agent do, and when should you use it?
Paca is a self-hosted, free, open-source project management platform designed to enable AI agents to collaborate as equal teammates with humans in a Scrum team. Key components include a React + TanStack Start frontend (apps/web), a Go + Gin REST API (services/api), a Node.js + Socket.IO real-time service (services/realtime), a Python + FastAPI + OpenHands SDK AI agent orchestration (services/ai-agent), and an MCP server (apps/mcp). The platform is configurable via files for workflows, statuses, field definitions, and more, and is extensible through WebAssembly (WASM) plugins for both backend and frontend. AI agents can be assigned to sprints, pick up tasks from the board, write BDD specs, and contribute to System Design Documents. Deployment is via a single Docker Compose command, and data is fully controlled by the user.
Paca runs a unified Scrumban board where humans and AI agents collaborate in real time. AI agents (such as Claude Code, Codex, Gemini CLI) can be assigned to sprints, pick up tasks, update statuses, and comment on the board, either through the ACP (Agent Client Protocol) bridge (paca-acp-bridge) or via the OpenHands SDK. The platform provides in-app AI chat for natural-language planning, a BDD collaboration editor for Gherkin scenarios co-authored by POs, BAs, and AI, and System Design Documents (SDD) to keep architecture visible. The MCP server (@paca-ai/paca-mcp) runs via npx and exposes tools for managing projects, tasks, sprints, documents, members, and more. The automation engine (v0.11.0) uses a visual Trigger → Condition → Action graph builder, with actions including task retargeting, AI agent dispatch, and external API calls. Plugin support includes WASM backend modules and frontend modules, installed via the Plugin Marketplace within the UI.
- A Scrum Master wants to integrate AI coding agents (e.g., Claude Code) as formal team members participating in sprint planning and task execution.
- A Product Owner uses in-app AI chat to turn requirements into epics and stories, refining them with BDD scenarios.
- A development team connects Claude Desktop via the MCP server to list sprints, create tasks, and add comments using natural language.
- A project manager needs a self-hosted, customizable board where workflows and fields can be adjusted via configuration files without coding.
- An AI agent (e.g., OpenHands) runs tasks in an isolated sandbox, updating the Kanban board via the Paca API without touching the host environment.
- A team wants to extend the platform via WASM plugins to add custom routes, data models, or UI components, with a capability-based permission model.
How do you install or deploy this agent?
To install Paca, you have two options:
- Interactive install script (recommended for production): Run the following command on a Linux server with Docker; no repository clone is required:
curl -fsSL https://github.com/Paca-AI/paca/releases/latest/download/install.sh | bashThe script walks you through configuration interactively and starts the full stack. For non-interactive (CI/scripts/AI agents), set PACA_YES=1 and use environment variables to steer other settings.
- Manual Docker Compose:
mkdir paca && cd paca
curl -fsSL https://github.com/Paca-AI/paca/releases/latest/download/docker-compose.yml -o docker-compose.yml
mkdir -p caddy
curl -fsSL https://github.com/Paca-AI/paca/releases/latest/download/Caddyfile -o caddy/Caddyfile
curl -fsSL https://github.com/Paca-AI/paca/releases/latest/download/.env.production.example -o .envGenerate secure passwords and secrets (POSTGRES_PASSWORD, ADMIN_PASSWORD, JWT_SECRET, ENCRYPTION_KEY, etc.), update .env, then run:
docker compose --env-file .env up -dOpen http://localhost and log in with the admin credentials you set.
How do you use this agent?
After installation, access the Paca UI via your browser and log in with the admin credentials. Create an API key in Settings → API Keys for MCP or ACP connections. To connect AI agents, use the MCP server: add a paca entry to your Claude Desktop config with command: npx, args: -y @paca-ai/paca-mcp, and PACA_API_KEY and PACA_API_URL environment variables. Alternatively, install the /paca skillset by running PACA_API_URL=http://localhost:8080 curl -fsSL https://raw.githubusercontent.com/Paca-AI/paca/master/scripts/install-paca-skills.sh | bash, which supports Claude Code, Gemini CLI, Cursor, among others. In daily use, create projects, sprints, and tasks, assign them to AI agents, and they will update statuses automatically. Customize workflows via configuration files and extend with plugins.
What are this agent's strengths and limitations?
- AI agents integrated as equal teammates, assignable on the board, not just chatbots
- Fully self-hosted; data remains on your infrastructure
- Free and open-source (Apache 2.0) with no per-seat costs
- Highly customizable via configuration files and WASM plugins with granular permission model
- Provides an MCP server compatible with any MCP client (e.g., Claude Desktop)
- Supports ACP (Agent Client Protocol), allowing integration with existing CLI agents (Claude Code, Codex, etc.) while preserving local credentials
- Relatively new project (v0.12.0), may lack long-term stability and features of mature PM tools
- Requires Docker and self-hosting infrastructure; not suitable for non-technical teams
- AI agent sandboxing requires additional resources (Docker containers)
- No managed cloud version unlike Jira/ClickUp; users must handle upgrades and security themselves
- AI agent behavior is dependent on the OpenHands SDK, which may limit customizability
- Plugin ecosystem is still nascent, community plugins may be limited
How does this agent compare with similar options?
Key facts side by side with the most closely related agents.
| Agent | Source review | Form / cost | Stars | Updated | Language | Full support on |
|---|---|---|---|---|---|---|
| Paca This agent | 53 · Major gaps | Self-hosted serviceFree | ★ 2k | 1d ago | Go | Codex · Claude Code |
| NextClaw | 58 · Major gaps | Desktop appFree + model costs | ★ 262 | today | TypeScript | Codex · Claude Code |
| PocketPaw | 46 · Major gaps | Desktop appFree + model costs | ★ 891 | 1mo ago | Python | Codex · OpenAI API · Claude API |
| CraftBot | 40 · Major gaps | CLIFree + model costs | ★ 396 | 13d ago | Python | OpenAI API · Claude API |
How does FollowAgents rate this agent?
Why each dimension lost points
Evidence shows: plugin system uses capability-based permission model, declares required host functions, and WASM sandbox restricts permissions; install script supports non-interactive mode, but user confirmation is only prompted during installation, not covering all operations; data flow transparency is limited, only mentions MCP and WebSocket, not detailing data flow; sensitive data handling has encryption keys and secret masking, but not comprehensive; dependency security has CI builds and dependency locking, but no vulnerability scanning evidence; external effects: AI agents run in sandbox, but external API calls not explicitly restricted; rollback: activity diff and revert available, but not covering all changes; source attribution: publisher identity not verified, but code and docs traceable. Deductions: user confirmation not comprehensive, data flow transparency insufficient, dependency security lacks vulnerability scanning, external effects control unclear.
Evidence shows: project structure clear, services have defined responsibilities, tests cover critical paths like ACP bridge and AI agent unit tests; dependency availability: uses Docker Compose and lock files, but no guarantee of dependency availability; failure messages: tests cover failure scenarios like send failure and interrupt, but not comprehensive. Deductions: dependency availability not fully proven, failure message coverage limited.
Evidence shows: targets Scrum teams, supports multiple AI agent integrations like Claude Code, Gemini CLI, scenarios clear; capability boundaries defined via plugin system and configuration, but limitations not detailed; trigger precision: automation engine supports multiple trigger types, but precision not detailed; environment fit: supports self-hosting and Docker Compose, but not all environments. Deductions: trigger precision not detailed, environment fit limited.
Evidence shows: information architecture clear, has architecture docs and guides; install notes detailed, multiple installation methods; naming stability: version releases and changelog, but naming conventions not explicit; examples and FAQ: examples and docs, but no FAQ; known limitations: not explicitly listed; license Apache 2.0, complete; versioning and changelog recorded; maintenance responsibility: contribution guide and security policy, but maintainers not explicit. Deductions: naming stability insufficient, known limitations not listed, maintenance responsibility unclear.
Evidence shows: output usability: MCP server and skills provided for integration; marginal value: as Jira alternative, offers AI collaboration features, unique value; cost-benefit: free open-source, self-hosted, low cost. Deductions: no evidence of actual usage effectiveness.
Evidence shows: claims consistent with code and docs, but no external verification; cross-source corroboration: tests and CI, but no independent verification; fact-inference separation: docs distinguish feature descriptions and design philosophy, but not explicitly labeled. Deductions: lack of external verification, fact-inference separation unclear.
- Publisher identity not verified, proceed with caution.
- Dependency security lacks vulnerability scanning evidence, recommend checking dependencies.
- User confirmation mechanism not comprehensive, some operations may proceed without confirmation.
- Data flow transparency limited, recommend reviewing data flow.
Recent releases
Latest entries from the project's GitHub Releases. Notes are quoted from upstream as written, not rewritten or tested by FollowAgents.
-
Release notes excerpt
What's Changed * feat(notification): include task/doc title in email event payload by @vhervatin in https://github.com/Paca-AI/paca/pull/553 * fix(automation): fire tag_added only when the tag is actually added (#554) by @laithalenooz in https://github.com/Paca-AI/paca/pull/556 * feat(authz): IAM-style policy roles replace flat permissions and rest… by @pikann in https://github.com/Paca-AI/paca/pull/557 * feat(authz): implement plugin access control and enhance permission v… by @pikann in https://github.com/Paca-AI/paca/pull/558 …View on GitHub ↗ -
Release notes excerpt
What's Changed * fix(web): save Timeline/board filters per user by default (PE-1) by @prakruthi2411 in https://github.com/Paca-AI/paca/pull/545 * fix(upgrade): don't exit silently when .env lacks AGENT_SERVER_IMAGE by @pikann in https://github.com/Paca-AI/paca/pull/549 * fix(users): use cursor-paginated server-side search in user and task pickers by @pikann in https://github.com/Paca-AI/paca/pull/551 * fix: make task, doc, folder, sprint and view updates atomic to prevent races by @pikann in https://github.com/Paca-AI/paca/pull/552 New Contributors …View on GitHub ↗ -
Release notes excerpt
What's Changed * feat(task): enhance partial updates for task fields and merge custom … by @pikann in https://github.com/Paca-AI/paca/pull/540 * feat: implement project export functionality by @pikann in https://github.com/Paca-AI/paca/pull/541 * feat(interaction): optimize dependency handling in InteractionLayout … by @pikann in https://github.com/Paca-AI/paca/pull/542 * feat(agent-dialog): require default folder for provider CLI agents by @pikann in https://github.com/Paca-AI/paca/pull/543 Full Changelog: https://github.com/Paca-AI/paca/compare/v0.18.4...v0.18.5 How to Upgrade …View on GitHub ↗