# AgentChat — LLM Chat + MCP Integration

> AgentChat is a modern LLM chat system (FastAPI + Vue) with knowledge base, tool calling, and MCP server integration, plus a Docker Compose deploy path.

## Install

Copy the content below into your project:

## Quick Use

```bash
git clone https://github.com/Shy2593666979/AgentChat.git
cd AgentChat

# per docker/README.md
cd docker
chmod +x start_linux.sh
./start_linux.sh

# UI: http://localhost:8090
# API: http://localhost:7860/docs
```

## Intro

AgentChat is a modern LLM chat system (FastAPI + Vue) with knowledge base, tool calling, and MCP server integration, plus a Docker Compose deploy path.

- **Best for:** Self-hosted chat workbenches that need RAG + tools + MCP in one stack
- **Works with:** Docker Compose; FastAPI backend; Vue 3 frontend; optional MySQL/Redis/MinIO services (per docker docs)
- **Setup time:** 30–90 minutes

## Practical Notes

- GitHub: 712 stars · 73 forks; pushed 2026-04-15 (verified via GitHub API).
- docker/README.md lists Docker 20.10+ and Compose 2.0+, plus default ports: UI 8090, API 7860, MySQL 3306, Redis 6379.
- docker/README.md uses a YAML config file (`docker_config.yaml`) instead of env vars for model keys and DB endpoints.

## Main

Operational hardening checklist:

1. Keep secrets out of git: store `docker_config.yaml` outside the repo or use a templated file + secret manager.
2. Put the API behind an auth layer (reverse proxy + JWT/OAuth) before exposing it to the internet.
3. Separate “chat UI” from “tool execution”: run tools in a constrained worker with clear timeouts and resource limits.
4. Add backups for MySQL and object storage from day one—RAG and chat history are your core assets.

Once stable, you can wire MCP servers as controlled integrations rather than ad-hoc scripts.

### FAQ

**Q: Is Docker required?**
A: No, but the repo includes a Docker deployment path that is the quickest way to trial the full stack.

**Q: Where do I configure model keys?**
A: docker/README.md says it uses a YAML config file (`docker_config.yaml`) rather than env vars.

**Q: What should I lock down first?**
A: Authentication + network exposure; then isolate tool execution and add backups.

## Source & Thanks

> Source: https://github.com/Shy2593666979/AgentChat
> License: MIT
> GitHub stars: 712 · forks: 84

---

<!-- ZH -->

## 快速使用

```bash
git clone https://github.com/Shy2593666979/AgentChat.git
cd AgentChat

# 按 docker/README.md
cd docker
chmod +x start_linux.sh
./start_linux.sh

# 前端：http://localhost:8090
# API：http://localhost:7860/docs
```

## 简介

AgentChat 是现代化 LLM 对话系统（FastAPI + Vue）：支持知识库检索、工具调用与 MCP server 集成，并提供一键 Docker Compose 部署，适合快速搭建可扩展的聊天工作台。

- **适合谁：** 需要把 RAG + 工具 + MCP 集成到一套自托管聊天栈的团队
- **可搭配：** Docker Compose；FastAPI 后端；Vue 3 前端；可选 MySQL/Redis/MinIO（见 docker 文档）
- **准备时间：** 30–90 分钟

## 实战建议

- GitHub：712 stars · 73 forks；最近更新 2026-04-15（GitHub API 验证）。
- docker/README.md 标注 Docker 20.10+ / Compose 2.0+；默认端口：前端 8090、API 7860、MySQL 3306、Redis 6379。
- docker/README.md 采用 `docker_config.yaml` 配置模型 key 与数据库端点，而不是环境变量。

## 主要内容

上线前建议做这些加固：

1. 密钥不进 git：把 `docker_config.yaml` 放到仓库外，或用模板 + 密钥管理系统注入。
2. 对外暴露前加鉴权：反向代理 + JWT/OAuth，避免裸奔 API。
3. 聊天 UI 与工具执行隔离：工具跑在受限 worker 中，设置超时与资源上限。
4. 从第一天就做备份：MySQL 与对象存储（RAG 索引与聊天历史是核心资产）。

稳定后再把 MCP servers 当作“受控集成”接入，而不是临时脚本。

### FAQ

**一定要 Docker 吗？**
答：不一定，但仓库提供 Docker 部署路径，是体验全栈最快的方式。

**模型 key 在哪配置？**
答：docker/README.md 表示用 `docker_config.yaml` 这类 YAML 配置，而不是环境变量。

**第一步先加固什么？**
答：先做鉴权与网络暴露控制，然后隔离工具执行并加上备份。

## 来源与感谢

> Source: https://github.com/Shy2593666979/AgentChat
> License: MIT
> GitHub stars: 712 · forks: 84


---
Source: https://tokrepo.com/en/workflows/agentchat-llm-chat-mcp-integration
Author: AI Open Source