GTPlanner: AI-Powered PRD Generator

<p align="center"> <img src="./assets/banner.png" width="800" alt="GTPlanner Banner"/> </p> <p align="center"> <strong>An intelligent Agent PRD generation tool that transforms natural language descriptions into structured technical documentation optimized for Vibe coding.</strong> </p> <p align="center"> <a href="#overview">Overview</a> • <a href="#web-ui-recommended">Web UI</a> • <a href="#claude-code-skill">Claude Code Skill</a> • <a href="#mcp-integration">MCP Integration</a> • <a href="#quick-start">Quick Start</a> • <a href="#configuration">Configuration</a> • <a href="#project-structure">Project Structure</a> • <a href="#prefab-ecosystem">Prefab Ecosystem</a> • <a href="#contributing">Contributing</a> • <a href="#license">License</a> </p> <p align="center"> <strong>Languages:</strong> <a href="README.md">English</a> • <a href="docs/zh/README.md">简体中文</a> • <a href="docs/ja/README.md">日本語</a> </p>

---

Overview—How to Make Agents Work for You?

• First, define the task: What are the inputs? What are the specific steps? What are the outputs? How do you define success? These are typically called SOPs. Work that can be SOPed can be automated by AI.
• Second, provide the right tools for your agent. Humans use Office suites, browse the web, manage files and data, etc. If you want agents to work, they need these tools too.
• Finally, specify how the agent should organize its outputs.

In the Agent context, there's a more appropriate term for this: context engineering. Specifically, before starting Code Agents (Claude Code/Cursor/Augment/Devin/...), we want to clearly define through documentation:

• design.md: Define what the work is.
• prefab.md: Define available tools and how to use them. We call these Prefabs.
• starter-kit: Define development frameworks and available environments (this part is mostly consistent across projects).

This is what GTPlanner does—simplifying the process of building Agents.

Why Choose GTPlanner

GTPlanner helps you easily generate an Agent PRD—a standard operating procedure (SOP) that AI can understand—to quickly automate your tasks. GTPlanner's design philosophy:

• Determinism: Eliminate AI ambiguity through clear SOPs (Agent PRDs), ensuring highly controllable and predictable execution paths and results.
• Composability: Break SOPs into reusable "Prefabs" and task modules, combining them like building blocks to create more complex workflows.
• Freedom: We don't lock you into execution platforms (like n8n), but instead use minimal AI frameworks and native Python code for maximum flexibility and freedom.

---

Web UI (Recommended)

For the best experience, we strongly recommend using the Web UI. It provides a streamlined AI planning workflow tailored for modern developers.

Key Advantages:

• Intelligent Planning Assistant: AI-assisted rapid generation of system architecture and project plans
• Instant Documentation: Automatically create comprehensive technical documentation
• Perfectly Suited for Vibe Coding: Optimized output for Cursor, Windsurf, GitHub Copilot
• Team Collaboration: Multi-format export for easy sharing

Try the Live Demo

---

Claude Code Skill

GTPlanner is available as a Claude Code plugin skill, enabling seamless PRD generation directly within Claude Code.

Trigger Words: PRD, 项目规划, 架构设计, 技术方案, 设计文档

Installation

1. Clone GTPlanner as a plugin:

cd ~/.claude/plugins
git clone https://github.com/OpenSQZ/GTPlanner.git

2. Configure environment:

cd GTPlanner
cp .env.example .env
# Edit .env with your LLM configuration

3. Initialize:

# In Claude Code, run:
/gtplanner-init

Usage

Simply mention trigger words in Claude Code:

帮我设计一个自动化视频处理流程：从 YouTube 下载视频，提取字幕，生成摘要

Or in English:

Design a workflow that monitors GitHub issues, analyzes sentiment, and auto-assigns labels

Available Tools

| Tool | Description | |------|-------------| | short_planning | Requirement analysis and scope definition | | tool_recommend | Technology stack recommendations | | research | Deep technical research (requires JINA_API_KEY) | | design | Generate design documents (quick/deep mode) |

Workflow

short_planning (initial) → tool_recommend → short_planning (technical) → design

Detailed skill documentation → Skill Guide

---

MCP Integration

<details> <summary>Click to expand MCP integration instructions</summary>

GTPlanner supports the Model Context Protocol (MCP) for direct use in AI programming tools.

<table> <tr> <td width="50%">

Using in Cherry Studio

</td> <td width="50%">

Using in Cursor

</td> </tr> </table>

Detailed configuration guide → MCP Documentation

</details>

---

Quick Start

Online Quick Experience (No Installation)

Try Web UI - WYSIWYG planning and documentation generation

Local Setup

Prerequisites

• Python ≥ 3.10 (3.11+ recommended)
• Package Manager: uv (recommended)
• LLM API Key: OpenAI / Anthropic / Azure OpenAI / Self-hosted compatible endpoint

Installation

git clone https://github.com/OpenSQZ/GTPlanner.git
cd GTPlanner

# Install dependencies with uv
uv sync

Configuration

Copy the configuration template and set your API Key:

cp .env.example .env
# Edit .env file and set required environment variables

Required Configuration:

LLM_API_KEY="your-api-key-here"
LLM_BASE_URL="https://api.openai.com/v1"
LLM_MODEL="gpt-5"

Detailed configuration guide (including common providers, Langfuse, etc.) → Configuration Documentation

CLI Usage

Interactive Mode

python gtplanner.py

After starting, simply input your requirements, for example:

Create a video analysis assistant that can automatically extract video subtitles, generate summaries and key points.

Direct Execution

python gtplanner.py "Design a document analysis assistant that supports PDF, Word document parsing and intelligent Q&A"

CLI detailed documentation (session management, parameter explanations, etc.) → CLI Documentation

API Usage

Start the FastAPI service:

uv run fastapi_main.py
# Runs on http://0.0.0.0:11211 by default

Visit http://0.0.0.0:11211/docs to view API documentation

API detailed documentation (endpoint descriptions, usage examples, etc.) → API Documentation

MCP Integration

cd mcp
uv sync
uv run python mcp_service.py

MCP detailed documentation (client configuration, available tools, etc.) → MCP Documentation

---

Configuration

GTPlanner supports multiple configuration methods:

• Environment Variables (.env file): API Key, Base URL, Model, etc.
• Configuration File (settings.toml): Language, tracing, vector services, etc.
• Langfuse Tracing (optional): Execution process tracing and performance analysis

Complete configuration guide → Configuration Documentation

---

Project Structure

GTPlanner/
├── README.md                  # Main documentation
├── gtplanner.py              # CLI entry point
├── fastapi_main.py           # API service entry
├── settings.toml             # Configuration file
│
├── .claude/                  # Claude Code plugin
│   ├── plugin.json          # Plugin manifest
│   ├── skills/gtplanner/    # → [Skill Guide](./.claude/skills/gtplanner/SKILL.md)
│   └── commands/            # Slash commands
│
├── agent/                   # Core agent code
│   ├── flows/              # Control flows
│   ├── subflows/           # Specialized subflows
│   ├── function_calling/   # Tool definitions
│   └── ...
│
├── utils/                  # Utilities
│
├── prefabs/                 # Prefab ecosystem
│   ├── README.md           # → [Prefab Documentation](./prefabs/README.md)
│   └── releases/           # Release management
│       ├── community-prefabs.json  # Prefab registry
│       └── CONTRIBUTING.md # → [Prefab Contributing Guide](./prefabs/releases/CONTRIBUTING.md)
│
├── mcp/                    # MCP service
│   └── README.md          # → [MCP Documentation](./mcp/README.md)
│
├── docs/                   # Documentation
│   ├── zh/                # Chinese documentation
│   ├── ja/                # Japanese documentation
│   ├── configuration.md   # Configuration guide
│   └── architecture/      # Architecture documentation
│
├── workspace/             # Runtime directory
│   ├── logs/             # Logs
│   └── output/           # Output documents
│
└── tests/                # Tests

System architecture documentation → Architecture Documentation

---

Prefab Ecosystem

GTPlanner extends capabilities through the Prefab ecosystem. Each Prefab is a standardized, reusable AI functional component.

What is a Prefab?

Prefabs are ready-to-use AI functional modules that can be:

• Discovered: GTPlanner automatically recognizes available Prefabs
• Deployed: Automatically deployed to the platform after PR merge
• Integrated: Called through standard APIs
• Version Managed: Semantic versioning

How Do Prefabs Enhance GTPlanner?

When you contribute a Prefab to community-prefabs.json:

1. Expand Planning Capabilities: GTPlanner learns about a new solution
2. Smart Recommendations: GTPlanner recommends appropriate Prefabs when generating plans
3. Automatic Integration: Planning documents include Prefab usage instr