Use Paca directly from Claude Code CLI with the /paca and /paca-setup slash commands. Once installed, Claude will use your Paca workspace for tasks, documentation, and sprint management — instead of creating local files.
Run the installer (works on macOS and Linux):
curl -fsSL https://raw.githubusercontent.com/Paca-AI/paca/master/scripts/install-claude-skill.sh | bashSecurity note: Review the script before running it —
curl | bashexecutes remote code directly. You can inspect it at the URL above, then runbash scripts/install-claude-skill.shfrom a local clone instead.
Or, from a local clone of this repo:
bash scripts/install-claude-skill.shThe installer copies two skill files to ~/.claude/commands/, making /paca and /paca-setup available in every Claude Code session.
The skill requires the Paca MCP server to be connected. After installing the skill, run /paca-setup inside a Claude Code session for an interactive setup walkthrough, or follow the quick steps below.
claude mcp add paca \
--env PACA_API_KEY=<your-api-key> \
--env PACA_API_URL=<your-paca-url> \
-- npx -y @paca-ai/paca-mcpReplace <your-api-key> (from Paca → Settings → API Keys) and <your-paca-url> (e.g. http://localhost:8080 or your hosted URL).
Create .claude/mcp.json in your project root:
{
"mcpServers": {
"paca": {
"command": "npx",
"args": ["-y", "@paca-ai/paca-mcp"],
"env": {
"PACA_API_KEY": "<your-api-key>",
"PACA_API_URL": "http://localhost:8080"
}
}
}
}Security: Do not commit API keys. Add
.claude/mcp.jsonto.gitignoreor injectPACA_API_KEYfrom your shell environment.
Add to the config file for your OS:
- macOS:
~/Library/Application Support/Claude/claude_desktop_config.json - Windows:
%APPDATA%\Claude\claude_desktop_config.json - Linux:
~/.config/Claude/claude_desktop_config.json
{
"mcpServers": {
"paca": {
"command": "npx",
"args": ["-y", "@paca-ai/paca-mcp"],
"env": {
"PACA_API_KEY": "<your-api-key>",
"PACA_API_URL": "http://localhost:8080"
}
}
}
}Restart Claude Desktop after saving.
All commands read your Paca documentation first — before taking any action, Claude calls list_documents and reads the relevant docs so it understands the project context. Documentation updates are always written back to Paca Docs via update_document or create_document, never as local files.
General-purpose Paca operations in plain English. Routes to the right tool based on intent.
/paca Fix the login redirect bug, assign to sprint 3
/paca What's in the current sprint?
/paca Mark task #42 as done
/paca ABC-17 is blocked — add a comment: needs design review
Converts requirements into a structured epic: creates the parent task, breaks it into child stories, and writes a spec document — all in Paca.
/paca-epic As a user I want to reset my password via email
/paca-epic #12 ← turn an existing requirement task into a full epic
Reads a task or document, identifies ambiguities (scope gaps, missing edge cases, undefined terms), asks targeted questions, then updates the spec in Paca with the resolved content.
/paca-clarify #42
/paca-clarify ABC-17
/paca-clarify "OAuth Integration Spec"
Decomposes a task or epic into smaller, independent, estimable sub-tasks and creates them in Paca.
/paca-breakdown #42
/paca-breakdown ABC-17
Plans a sprint: reads the backlog and project roadmap, recommends a task set that fits stated capacity, then assigns tasks to the sprint and sets the sprint goal.
/paca-sprint
/paca-sprint next sprint, 30 points capacity
/paca-sprint sprint 4, goal: ship the auth flow
Estimates story points for one or more tasks using the Fibonacci scale, with reasoning, then writes the estimates back to the tasks.
/paca-estimate #42
/paca-estimate #42 #43 #44
/paca-estimate ← estimates all unestimated tasks in the current sprint
Scores tasks by business value, urgency, effort, and dependencies against the project roadmap, then updates their priority fields.
/paca-prioritize
/paca-prioritize #42 #43 #44
Executes a task end-to-end: marks it in progress, reads all relevant docs, does the work (code, writing, research), then marks it done and updates any affected Paca documentation.
/paca-do #42
/paca-do ABC-17
Derives test cases from acceptance criteria, runs them, and posts results as a task comment. Advances the task status on pass, reverts to in-progress on fail.
/paca-test #42
/paca-test ABC-17
Writes or updates a document in Paca Docs. Reads existing docs first to match tone and avoid duplication.
/paca-doc #42 ← document the feature in task #42
/paca-doc "API Authentication Guide" ← create a new guide
/paca-doc ABC-17 update ← update an existing doc
Interactive setup wizard. Walks you through connecting Claude Code to your Paca instance and verifying the connection.
/paca-setup
| Variable | Required | Default | Description |
|---|---|---|---|
PACA_API_KEY |
Yes | — | API key (Paca → Settings → API Keys) |
PACA_API_URL |
No | http://localhost:8080 |
Your Paca instance URL |
To make Claude always prefer Paca tools in a project (without needing to type /paca every time), add this to your project's CLAUDE.md:
## Project management
This project uses Paca for all project management. When working in this codebase:
- **Tasks and to-dos** → use `create_task` / `list_tasks` via the Paca MCP tools. Do not create local TODO files or add TODO comments.
- **Documentation** → use `create_document` / `update_document` via Paca MCP. Do not create standalone `.md` docs unless they belong in the repository (e.g. README, CONTRIBUTING).
- **Sprint planning** → use `create_sprint` / `list_sprints` via Paca MCP.
If Paca MCP tools are not available, say so and ask the user to run `/paca-setup`.rm ~/.claude/commands/paca.md ~/.claude/commands/paca-setup.mdSee mcp-server-setup.md for the complete tool reference.