Pagent

CLI tool that orchestrates Claude Code agents to transform PRDs into specs and working code.

flowchart LR
    PRD[PRD.md] --> A[architect]
    A --> Q[qa]
    A --> S[security]
    Q & S --> I[implementer]
    I --> V[verifier]
    V --> O[outputs/]

    subgraph Specs
        A
        Q
        S
    end

    subgraph Code
        I
        V
    end

Loading

Quick Start

brew install tuannvm/mcp/pagent
pagent ui

The TUI guides you through selecting a PRD, persona, and running agents:

 ██████╗  █████╗  ██████╗ ███████╗███╗   ██╗████████╗
 ██╔══██╗██╔══██╗██╔════╝ ██╔════╝████╗  ██║╚══██╔══╝
 ██████╔╝███████║██║  ███╗█████╗  ██╔██╗ ██║   ██║
 ██╔═══╝ ██╔══██║██║   ██║██╔══╝  ██║╚██╗██║   ██║
 ██║     ██║  ██║╚██████╔╝███████╗██║ ╚████║   ██║
 ╚═╝     ╚═╝  ╚═╝ ╚═════╝ ╚══════╝╚═╝  ╚═══╝   ╚═╝

 From idea to implementation, orchestrated.

 Input                       Persona
 > examples/sample-prd.md    > Balanced - Standard

 Output                      Action
 > ./outputs                 > ▶ Run

 ↑ up · ↓ down · / filter · enter select

Or use the CLI directly:

pagent run ./prd.md --sequential -v

Prerequisites

• Claude Code installed and authenticated

Installation

Homebrew (recommended):

brew install tuannvm/mcp/pagent

Alternative installation methods

Binary download:

# Detect OS and architecture automatically
curl -sSL https://github.com/tuannvm/pagent/releases/latest/download/pagent_$(uname -s | tr '[:upper:]' '[:lower:]')_$(uname -m | sed 's/x86_64/amd64/').tar.gz | tar xz
sudo mv pagent /usr/local/bin/

From source (requires Go 1.21+):

git clone https://github.com/tuannvm/pagent && cd pagent && make install

Commands

Command	Description
pagent run <prd>	Run agents on PRD
pagent ui [prd]	Interactive dashboard
pagent status	Check running agents
pagent logs <agent>	View agent output
pagent message <agent> "msg"	Send guidance
pagent stop [--all]	Stop agents
pagent init	Create config file
pagent mcp	Run as MCP server

Common Options

pagent run prd.md --agents architect,qa   # Run specific agents
pagent run prd.md --sequential            # Run in dependency order
pagent run prd.md --resume                # Skip up-to-date outputs
pagent run prd.md --output ./docs/        # Custom output directory
pagent run prd.md --persona minimal       # Use minimal persona

Agents

Agent	Output	Role
architect	architecture.md	System design, API, data models
qa	test-plan.md	Test strategy and cases
security	security-assessment.md	Threat model, mitigations
implementer	code/*	Complete codebase
verifier	code/*_test.go	Tests + verification

Configuration

Run pagent init to create .pagent/config.yaml. Key options:

• persona: minimal | balanced | production
• preferences: API style, testing depth, language
• stack: Cloud, database, CI/CD choices

See docs/tutorial.md for full config reference.

MCP Server

Pagent can run as an MCP (Model Context Protocol) server, enabling integration with Claude Desktop, Claude Code, and other MCP clients.

# Stdio transport (default) - for Claude Desktop
pagent mcp

# HTTP transport - for web integration
pagent mcp --transport http --port 8080

# HTTP with OAuth 2.1 authentication
pagent mcp --transport http --oauth --issuer https://company.okta.com --audience api://pagent

Available MCP Tools:

Tool	Description
run_agent	Run a single agent on a PRD
run_pipeline	Run the full agent pipeline
list_agents	List available agents
get_status	Check running agent status
send_message	Send guidance to a running agent
stop_agents	Stop running agents

See docs/tutorial.md for setup instructions.

Claude Code Skill

This repo includes a Claude Code skill (.claude/skills/pagent/SKILL.md) that teaches Claude how to use pagent. When you open this project in Claude Code, it automatically knows pagent commands and workflows.

Documentation

Doc	Content
Tutorial	Step-by-step usage guide
Architecture	Technical design and internals
Roadmap	Future plans

Troubleshooting

"timeout waiting for agent" - Check claude --version, increase --timeout 600

"port already in use" - Run pagent stop --all

See docs/tutorial.md#troubleshooting for more.

Contributing

See CONTRIBUTING.md.

Roadmap

Currently supports Claude Code only. Coming next:

• Multi-LLM support - Gemini, Codex, AMP
• Simplified config - Fewer options, smarter defaults
• Better UX - Guided setup, clearer outputs

Recently shipped:

• MCP Server - Integrate with Claude Desktop and MCP clients

See full roadmap.

Acknowledgments

This project is built on AgentAPI by Coder - a brilliant HTTP wrapper for Claude Code that makes agent orchestration possible. Pagent wouldn't exist without it.

License

MIT