go-mcp-demo
About
go-mcp-demo is an MCP Server and Client demo built on mark3labs/mcp-go that integrates with Ollama to enable local large language models to call tools via natural language. It targets developers who want to see how an MCP server communicates with an MCP client and performs tasks…
Details
- Author
- Patrick7241
- GitHub stars
- 4
- Downloads
- 129
- Categories
- Other
Jump to
- Integrates MCP Server and Client communication framework
- Supports reading local .sql files via natural language
- Supports executing arbitrary MySQL queries via natural language
- Uses Ollama local LLM for inference (default port localhost:11434)
Install Ollama and pull a model (e.g., qwen2:1.5b). Edit pkg/config/config.yml to set the Ollama model, MySQL connection, and a .sql file path. Then run cmd/server/main.go to start the main server (default port 2001), and separately run cmd/mcp-server/main.go to start the MCP server (default port 2002). The MCP server registers two tools: read_file and query_db.
go-mcp-demo
本项目基于 mark3labs/mcp-go 实现了一个 MCP Server 与 MCP Client 通信的演示,结合 Ollama 启动本地大语言模型,实现自然语言调用工具的能力。
✨ 功能概览
- 集成 MCP Server + Client 通信框架
- 支持基于自然语言执行:
- 读取本地 .sql 文件
- 执行任意 MySQL 查询
- 使用 Ollama 本地大模型进行推理(默认端口:http://localhost:11434)
---
🚀 快速开始
1. 安装 Ollama
请参考官网:https://ollama.com
根据操作系统下载安装 Ollama CLI 工具。
---
2. 拉取并运行模型
在命令行中执行以下命令,拉取你需要的模型(示例使用 qwen2:1.5b):
ollama pull qwen2:1.5b
ollama run qwen2:1.5b
模型将监听本地端口:
http://localhost:11434/api/generate
---
3. 修改配置文件
打开并编辑
pkg/config/config.yml:
ollama:
model: qwen2:1.5b
mysql:
host: localhost
port: 3306
user: root
password: your_password
database: your_database
sql:
sqlFilePath: E:/your/path/to/test.sql
> ✅ 注意事项:
>
>
model 名称需与 ollama pull 和 ollama run 保持一致
> mysql 信息替换为你本地或远程数据库连接参数
> sqlFilePath 设置为你希望读取的 .sql 文件绝对路径
---
4. 启动主服务(Client)
进入主服务目录并运行:
cd cmd/server
go run main.go
默认监听地址:
http://localhost:2001
---
5. 启动 MCP 服务(Server)
在新终端中启动 MCP 服务:
cd cmd/mcp-server
go run main.go
默认监听地址:
http://localhost:2002
服务将自动注册以下工具:
read_file: 读取指定的 .sql 文件内容
query_db`: 执行任意 SQL 并返回查询结果---
Sign in to leave a review
Use Google, GitHub, or an email account so ratings stay tied to real people.
No reviews posted yet.



