baby-gpt
Provides web search capabilities using DuckDuckGo as a backend.
Integrates with GitHub via MCP, allowing management of repositories, issues, pull requests, and code search.
Optionally enables image generation using OpenAI's API.
Click on "Deploy Server".
Wait a few minutes for the server to deploy. Once ready, it will show a "Started" state.
In the chat, type
@followed by the MCP server name and your instructions, e.g., "@baby-gptSearch the web for today's top tech news"
That's it! The server will respond to your query, and you can continue using it as needed.
Here is a step-by-step guide with screenshots.
baby-gpt — 通用 ReAct Agent
从零实现的通用 Agent,完整覆盖 ReAct 架构的每一个环节。 Hello Agents 课程毕业设计项目。
快速开始
方式一:本地运行(开发者)
前置条件:Python 3.10+,Node.js 18+(MCP 功能需要)
# 1. 克隆仓库
git clone <repo-url>
cd baby-gpt
# 2. 安装依赖
make install
# 或者手动:python3 -m venv venv && venv/bin/python3 -m pip install -r requirements.txt
# 3. 配置 API Key
cp .env.example .env
# 编辑 .env,填入你的 DEEPSEEK_API_KEY
# 4. 启动
make run方式二:Docker 运行(推荐分发给他人)
前置条件:Docker
# 1. 构建镜像
make docker-build
# 2. 配置 Key(只需做一次)
cp .env.example .env
# 编辑 .env
# 3. 运行
make docker-runRelated MCP server: OpenClaw MCP Server
配置文件
复制 .env.example 为 .env,填入以下 Key:
变量名 | 必填 | 说明 |
| 可选 | LLM 提供方: |
| ✅* | DeepSeek API Key(默认主 LLM;用 Claude 时非必填) |
| ✅* | Claude API Key( |
| 可选 | Claude 模型,默认 |
| 可选 | 图片生成功能需要 |
| 可选 | 启用 GitHub MCP(26 个工具) |
| 可选 | 更好的网络搜索,无则自动用 DuckDuckGo |
功能
双 LLM 后端 — DeepSeek(ReAct 文本,默认)/ Claude(Anthropic SDK 原生
tool_use结构化调用),LLM_PROVIDER一键切换ReAct 核心循环 — Thought → Action → Observation 迭代
工具注册系统 —
@register_tool装饰器,tools/目录自动扫描远程 Skill 加载 —
find_skills+load_skill,运行时扩展能力能力自扩展 — 遇到未知任务自动搜索并安装工具
多轮对话 —
AgentSession滑动窗口,保留最近 20 轮上下文网络搜索 — DuckDuckGo(免费)/ Tavily(可选)双后端
GitHub MCP — 连接 GitHub MCP Server,读写仓库、搜索代码、管理 Issue/PR
RAG 记忆 — Chroma 向量数据库,
remember/recall工具MCP Server — 把自己的工具暴露给 Claude Desktop / Cursor
项目结构
baby-gpt/
├── core/
│ ├── agent.py ← ReAct 主循环
│ ├── llm_client.py ← LLM 调用(DeepSeek / OpenAI / Claude·Anthropic SDK)
│ ├── registry.py ← 工具注册中心
│ ├── session.py ← 多轮对话管理
│ └── tracer.py ← 可观测性(Timeline + 日志)
├── tools/
│ ├── shell.py ← Shell 执行(黑名单安全模型)
│ ├── filesystem.py ← 文件读写
│ ├── search.py ← 网络搜索
│ ├── memory.py ← RAG 记忆
│ ├── utils.py ← 计算器等工具
│ └── mcp_client.py ← MCP Client(连接外部 MCP Server)
├── skills/
│ └── remote_loader.py ← find_skills + load_skill
├── mcp_server/
│ └── server.py ← MCP Server(暴露给 Claude Desktop)
├── main.py ← 启动入口
├── Dockerfile
├── Makefile
└── .env.example接入 Claude Desktop(MCP Server 模式)
在 ~/Library/Application Support/Claude/claude_desktop_config.json 添加:
{
"mcpServers": {
"baby-gpt": {
"command": "/path/to/baby-gpt/venv/bin/python",
"args": ["/path/to/baby-gpt/mcp_server/server.py"]
}
}
}或者 Docker 方式:
{
"mcpServers": {
"baby-gpt": {
"command": "docker",
"args": ["run", "-i", "--rm", "-v", "/path/to/.env:/app/.env", "baby-gpt", "python", "mcp_server/server.py"]
}
}
}This server cannot be deployed
Maintenance
Related MCP Connectors
- QuallaaOAuthcom.quallaa
Talk to your public-facing AI from any MCP client — Claude, ChatGPT, Cursor, Cline, Windsurf.
Let ChatGPT, Claude & Cursor use your Mac: email, calendar, iMessage, Teams, files. Local, free.
Your apps, skills, MCP servers and keys from ahel.ai, served to Claude, ChatGPT, Cursor and Codex.
Persistent memory for Claude Code and Cursor. Stop re-explaining your project every session.
Related MCP Servers
- AlicenseAqualityCmaintenanceProvides filesystem, web search, SQLite, and system tools for AI assistants like Claude, enabling secure access to local resources and the web.6MIT
- AlicenseNot gradedqualityDmaintenanceExposes your OpenClaw assistant to Claude Desktop via Streamable HTTP transport, enabling interaction between the two AI systems.6 npm1MIT
- AlicenseNot gradedqualityDmaintenanceTurns any CLI tool or REST API into an MCP server for Claude, enabling Claude to use git, databases, or any API through natural language.MIT
- AlicenseAqualityDmaintenanceExposes local file I/O, a restricted shell whitelist, and SQLite helpers for safe integration with AI clients like Cursor and Claude Desktop.7MIT