AutoLearn 是一个模型上下文协议（MCP）服务器，允许 AI 代理从自然语言动态创建和重用新的技能（功能代码工作流）。

与静态MCP服务器不同，AutoLearn引入了一个编码代理，该代理：

• 将推理痕迹转换为晶体化内存（Python函数）。
• 自动决定要确定哪些工作流程。
• 动态更新其MCP规格，使消费代理能够立即使用新技能。

了解更多：https://www.autolearn.dev/

该项目包含一个前端演示应用，用户可以：

• 与使用自动学习的消费代理聊天。
• 参见自动生成的Python代码以获取新技能。
• 实时查看更新后的MCP规范。
• 交互式执行技能。

---

特色

• 动态技能创建：通过OpenAI集成实现自然语言→Python代码工作流程
• 结晶存储器：常用或复杂推理，作为可执行代码保存
• MCP 服务器：通过 HTTP 传输实现完整的 JSON-RPC 2.0 协议兼容性
• 实时更新：WebSocket事件用于skill_added、skill_executed、mcp_updated

AutoLearn

AutoLearn is a Model Context Protocol (MCP) server that lets AI agents dynamically create and reuse new skills (functional code workflows) from natural language.

Unlike static MCP servers, AutoLearn introduces a coding agent that:

• Converts reasoning traces into crystalized memory (Python functions).
• Decides automatically which workflows to crystalize.
• Updates its MCP spec dynamically so consuming agents can use new skills immediately.

Learn More: https://www.autolearn.dev/

The project includes a frontend demo app where users can:

• Chat with a consuming agent that uses AutoLearn.
• See auto-generated Python code for new skills.
• View the updated MCP spec in real time.
• Execute skills interactively.

---

Features

• Dynamic Skill Creation: Natural language → Python code workflows via OpenAI integration
• Crystalized Memory: Frequently used or complex reasoning preserved as executable code
• MCP Server: Full JSON-RPC 2.0 protocol compliance over HTTP transport
• Real-Time Updates: WebSocket events for skill_added, skill_executed, mcp_updated
• Persistence: SQLite database for skills, sessions, and operational data
• Frontend Demo: T3 stack with chat, skill viewer, MCP spec viewer, and execution panel
• Comprehensive Testing: 54/54 tests passing with full integration coverage

---

Tech Stack

• Backend: Python 3.11+, FastAPI, OpenAI API
• Frontend: T3 Stack (Next.js, TypeScript, Tailwind CSS, tRPC) with shadcn/ui components
• Testing: Pytest (backend), Vitest (frontend)

---

Repository Structure

autolearn/
├── backend/          # FastAPI MCP server + skill engine
│   ├── app.py        # FastAPI application
│   ├── schemas.py    # Pydantic models
│   ├── skill_engine.py # Skill registry and execution
│   └── openai_client.py # OpenAI integration
├── frontend/         # T3 Stack frontend (Next.js, TypeScript, Tailwind, tRPC)
│   ├── src/          # Source code
│   │   ├── components/   # UI components using shadcn/ui
│   │   ├── pages/    # Next.js pages
│   │   └── server/   # tRPC router definitions
├── tests/            # Unit and integration tests
├── docs/             # Documentation (PRD, design notes)
├── skills.db         # SQLite database for skill persistence
└── README.md

---

Getting Started

Prerequisites

• Python 3.11+
• OpenAI API key

1. Clone the Repository

git clone https://github.com/tarkaai/autolearn.git
cd autolearn

2. Setup Backend and Frontend

python -m venv .venv
source .venv/bin/activate   # or .venv\Scripts\activate on Windows
pip install -e .   # Install in development mode

# Navigate to the frontend directory
cd frontend

# Install dependencies
npm install