MCP Data Service · MCP 数据服务

Expose data query + auto-insights (anomaly detection) as an MCP Server — plug into Claude Desktop, Cursor, or any MCP client in 30 seconds. 把「数据查询 + 自动洞察」包装成 MCP 标准工具，Claude Desktop / Cursor 即插即用——给任意 AI Agent 装上数据能力，30 秒接入。

---

English · 中文

English

What is this?

An MCP (Model Context Protocol) server that turns a data table into callable tools for any AI agent. No LLM API needed — everything is computed locally with pandas, so it is free and fast.

Tools

Tool	What it does
query_video_stats(metric, rule, condition)	Query by metric: TOP N / BOTTOM N / AVG / SUM / by date / with filter
data_insights()	Auto insights: overview + anomaly detection (1.5σ) + recommendations
list_columns()	List all column names and types

Quick Start

pip install -r requirements.txt

# Self-test (direct calls + MCP handshake)
python tests/test_server.py

# Run the server (stdio transport; clients below will launch it automatically)
python server.py

Connect from Claude Desktop / Cursor

Add to your MCP config (e.g. claude_desktop_config.json):

{
  "mcpServers": {
    "data-service": {
      "command": "python",
      "args": ["/absolute/path/to/mcp-data-service/server.py"]
    }
  }
}

Restart the client, then ask your agent: "What's the video with the highest completion rate?" or "Run auto insights on the data."

Project Layout

server.py          FastMCP server (@mcp.tool registers 3 tools)
data/video_stats.csv   Sample data
tests/test_server.py   Direct-call + MCP-stdio handshake tests
requirements.txt   fastmcp + pandas

Tests

python tests/test_server.py
# [PASS] query_video_stats direct calls
# [PASS] data_insights auto insights
# [PASS] MCP stdio handshake: list_tools + call_tool

Roadmap

• [ ] Publish to PyPI (pipx install mcp-data-service)
• [ ] SQLite / PostgreSQL backend
• [ ] Multiple data sources per server

⬆ Back to top

---

中文

这是什么？

一个 MCP（Model Context Protocol） Server：把一张数据表变成任何 AI Agent 可调用的工具。完全不依赖 LLM API——全部用 pandas 本地计算，免费且快速。

工具清单

工具	作用
query_video_stats(metric, rule, condition)	按指标查询：TOP N / BOTTOM N / 平均值 / 总和 / 按日期 / 带条件过滤
data_insights()	自动洞察：数据总览 + 异常下探（1.5σ）+ 运营建议
list_columns()	列出全部列名与类型

快速开始

pip install -r requirements.txt

# 自测（直接调用 + MCP 握手）
python tests/test_server.py

# 启动（stdio 传输，客户端按下面配置自动拉起）
python server.py

接入 Claude Desktop / Cursor

在客户端 MCP 配置（如 claude_desktop_config.json）里加：

{
  "mcpServers": {
    "data-service": {
      "command": "python",
      "args": ["/绝对路径/mcp-data-service/server.py"]
    }
  }
}

保存并重启客户端，即可让 Agent 直接调用「完播率最高的视频」「自动洞察数据异常」等能力。

文件结构

server.py          FastMCP Server（@mcp.tool 注册 3 个工具）
data/video_stats.csv   示例数据
tests/test_server.py   直调 + MCP stdio 握手双测试
requirements.txt   fastmcp + pandas

测试

python tests/test_server.py
# [PASS] query_video_stats 直接调用
# [PASS] data_insights 自动洞察
# [PASS] MCP stdio 握手：list_tools + call_tool

开发计划

• [ ] 发布 PyPI（pipx install mcp-data-service）
• [ ] SQLite / PostgreSQL 后端
• [ ] 单个 Server 支持多个数据源

⬆ 返回顶部

---

License · 许可证

MIT © lengzhanbao