Paca

Paca is a self-hosted, open-source project management platform built around the idea that AI agents should participate in the Scrum process as genuine teammates — not as chatbot add-ons bolted to the side. Released under the Apache 2.0 license, it runs on a single Docker Compose command and keeps all data on your own infrastructure. The project reached v0.4.5 as of June 2026 and has accumulated roughly 980 GitHub stars since its March 2026 creation.

What It Is

Paca is a Scrumban board and sprint management tool designed for mixed human-AI teams. Where conventional tools like Jira, Trello, ClickUp, and Monday treat AI as a peripheral automation layer, Paca assigns AI agents to sprints, surfaces them on the same board as human teammates, and lets them pick up tasks, write BDD specs, and update status in real time. The platform is structured around a four-phase P·A·C·A cycle — Plan, Act, Check, Adapt — which mirrors both Scrum and the scientific method, and is grounded in the Cynefin framework's recognition that complex work requires adaptive teams rather than rigid pipelines.

Architecture and Stack

Paca ships as a multi-service stack with clearly separated concerns:

• apps/web — React + TanStack Start + shadcn/ui frontend
• services/api — Go + Gin core business logic and REST API
• services/realtime — Node.js + Socket.IO for real-time event fan-out
• services/ai-agent — Python + FastAPI + OpenHands SDK for AI agent orchestration; each agent runs in its own isolated sandbox container
• apps/mcp — the @paca-ai/paca-mcp MCP server published on npm
• PostgreSQL for persistent storage, Valkey for cache and async event streams

The plugin system compiles extensions to WebAssembly (WASM) for the backend — supporting Go, Rust, AssemblyScript, or anything with a WASM target — and standard module bundles for the frontend. Plugins run in a sandboxed, capability-scoped environment and can be installed from a built-in Plugin Marketplace without touching the command line.

MCP Server and Claude Code Integration

Paca ships a first-class MCP (Model Context Protocol) server that gives any compatible AI agent structured access to projects, tasks, sprints, documents, members, and custom fields. The server is published as @paca-ai/paca-mcp on npm and connects to Claude Desktop, Claude Code, or any MCP-compatible client via a short JSON configuration block.

For Claude Code users, Paca provides a set of slash-command skills — /paca, /paca-epic, /paca-sprint, /paca-do, /paca-test, and more — installed globally with a single curl command. Each skill reads project documentation before acting, keeping agent output grounded in how the product actually works.

Update: v0.4.0 and v0.4.5

The v0.4.0 release introduced two headline features: in-app AI chat at the project level (create or update epics, stories, tasks, and docs in plain English without leaving Paca) and activity diff & revert (every field change shows a before/after diff; one click rolls it back). The latest release, v0.4.5, was published on June 16, 2026, indicating active development cadence. The repository's roadmap and releases pages track ongoing direction.

Self-Hosted Deployment Model

Paca is designed to run entirely on your own infrastructure. The recommended path is a single curl | bash installer that walks through configuration interactively and starts the full stack. A manual Docker Compose path is also documented for teams that want explicit control. Optional services — PostgreSQL, MinIO, and the AI agent container — can each be scaled to zero and replaced with external equivalents (e.g., AWS S3, an external database). Database migrations run automatically on API startup, and upgrades require only docker compose pull followed by docker compose up -d.