📌Overview

Purpose: The XiYan MCP Server facilitates interaction with databases through natural language queries, powered by advanced text-to-SQL capabilities.

Overview: The XiYan MCP Server implements a Model Context Protocol (MCP) that enables users to seamlessly perform database queries using natural language. Leveraging the XiYan-SQL framework, it showcases state-of-the-art performance in text-to-SQL applications and provides flexible integration options, including local and remote modes.

Key Features:

Natural Language Data Retrieval: Users can query databases using natural language, which the system converts to SQL queries, simplifying the interaction with complex data systems.
Support for Multiple Database Systems: The server supports popular database dialects, specifically MySQL and PostgreSQL, offering flexibility in deployment.
Local Mode for Enhanced Security: A pure local mode option is available, allowing users to operate without an API key, thereby enhancing security and privacy.
Integration with Advanced LLMs: It supports both general Large Language Models (LLMs) and specialized state-of-the-art models, enabling diverse applications based on user needs.

XiYan MCP Server

A Model Context Protocol (MCP) server that enables natural language queries to databases
powered by XiYan-SQL, SOTA of text-to-sql on open benchmarks

[English] | 中文
Follow me on Weibo

Features
Preview
- Architecture
- Best Practice
- Tools Preview
Installation
- Installing from pip
- Installing from Smithery.ai
Configuration
- LLM Configuration
  - General LLMs
  - Text-to-SQL SOTA model
  - Local Model
- Database Configuration
  - MySQL
  - PostgreSQL
Launch
- Claude Desktop
- Cline
- Goose
- Cursor
- Witsy
It Does Not Work
Citation

Features

🌐 Fetch data by natural language through XiYanSQL
🤖 Support general LLMs (GPT, qwenmax), Text-to-SQL SOTA model
💻 Support pure local mode (high security!)
📝 Support MySQL and PostgreSQL
🖱️ List available tables as resources
🔧 Read table contents

Preview

Architecture

There are two integration modes for this server:

Remote mode (default): Requires an API key to access the xiyanSQL-qwencoder-32B model from a service provider (see Configuration).
Local mode: More secure, does not require an API key.

Best Practice and Reports

Evaluation on MCPBench

XiYan MCP server outperforms MySQL MCP and PostgreSQL MCP servers by 2-22 percentage points on MCPBench. Detailed results: MCPBench, report Evaluation Report on MCP Servers.

Tools Preview

get_data: Natural language interface to query databases. Converts natural language to SQL using a built-in model and returns query results.
{dialect}://{table_name}: Retrieves sample data from a specified table for model reference.
{dialect}://: Lists current database names.

Installation

Installing from pip

Requires Python 3.11+.
Install with pip:

pip install xiyan-mcp-server

Run the server:

python -m xiyan_mcp_server

To use, create a YAML config file and then run:

env YML=path/to/yml python -m xiyan_mcp_server

Installing from Smithery.ai

Visit @XGenerationLab/xiyan_mcp_server (not fully tested).

Configuration

Use a YAML config file to configure the server. Example config_demo.yml:

model:
  name: "XGenerationLab/XiYanSQL-QwenCoder-32B-2412"
  key: ""
  url: "https://api-inference.modelscope.cn/v1/"

database:
  host: "localhost"
  port: 3306
  user: "root"
  password: ""
  database: ""

LLM Configuration

Version	General LLMs (GPT, qwenmax)	SOTA model by Modelscope	SOTA model by Dashscope	Local LLMs
Description	basic, easy to use	best performance, stable, recommended	best performance, for trial	slow, high-security
Name	official model name (e.g. gpt-3.5-turbo)	XGenerationLab/XiYanSQL-QwenCoder-32B-2412	xiyansql-qwencoder-32b	xiyansql-qwencoder-3b
Key	API key from provider (OpenAI, Alibaba)	API key of Modelscope	API key via email	""
URL	provider endpoint (e.g. OpenAI API URL)	https://api-inference.modelscope.cn/v1/	https://xiyan-stream.biz.aliyun.com/service/api/xiyan-sql	http://localhost:5090

General LLMs

Example config for GPT-3.5:

model:
  name: "gpt-3.5-turbo"
  key: "YOUR KEY"
  url: "https://api.openai.com/v1"
database:

For Alibaba Qwen-max:

model:
  name: "qwen-max"
  key: "YOUR KEY"
  url: "https://dashscope.aliyuncs.com/compatible-mode/v1"
database:

Text-to-SQL SOTA model

Recommended: XiYanSQL-qwencoder-32B (GitHub), SOTA on Bird benchmark.

Two usage options: Modelscope or Alibaba Cloud DashScope.

(1) Modelscope version

Get API key from Modelscope (API docs):

model:
  name: "XGenerationLab/XiYanSQL-QwenCoder-32B-2412"
  key: ""
  url: "https://api-inference.modelscope.cn/v1/"

(2) Dashscope version

Set environment variables with a key requested by email to godot.lzl@alibaba-inc.com:

name: "YOUR NAME"
email: "YOUR EMAIL"
organization: "Your organization"

Example config:

model:
  name: "xiyansql-qwencoder-32b"
  key: "KEY"
  url: "https://xiyan-stream.biz.aliyun.com/service/api/xiyan-sql"
database:

Note: This service is for trial only. Contact for production use. You can also deploy the model yourself.

Local Model

Slow (~12 seconds per query). For a stable and fast service, use Modelscope version.

Requirements:

PC/Mac with at least 16GB RAM
6GB disk space

Steps:

Install dependencies:

pip install flask modelscope torch==2.2.2 accelerate>=0.26.0 numpy=2.2.3

(Optional) Download model manually:

modelscope download --model XGenerationLab/XiYanSQL-QwenCoder-3B-2502

Run local server:

python local_xiyan_server.py

Server runs at http://localhost:5090/.

Sample config:

model:
  name: "xiyansql-qwencoder-3b"
  key: "KEY"
  url: "http://127.0.0.1:5090"

Database Configuration

Connection info: host, port, user, password, database.

Supports MySQL and PostgreSQL.

MySQL

database:
  host: "localhost"
  port: 3306
  user: "root"
  password: ""
  database: ""

PostgreSQL

Install driver:

pip install psycopg2

Config example:

database:
  dialect: "postgresql"
  host: "localhost"
  port: 5432
  user: ""
  password: ""
  database: ""

Launch

Claude Desktop

Add to Claude Desktop config:

{
    "mcpServers": {
        "xiyan-mcp-server": {
            "command": "python",
            "args": ["-m", "xiyan_mcp_server"],
            "env": {"YML": "PATH/TO/YML"}
        }
    }
}

Cline

Same config as Claude Desktop.

Goose

Run with:

env YML=path/to/yml python -m xiyan_mcp_server

Cursor

Use same command as Goose.

Witsy

Command:

python -m xiyan_mcp_server

Add environment variable key YML with value as the path to your yml file.

It Does Not Work!

Contact:
Ding Group钉钉群
Follow on Weibo

Citation

If you find our work helpful, please cite:

@article{xiyansql,
      title={A Preview of XiYan-SQL: A Multi-Generator Ensemble Framework for Text-to-SQL}, 
      author={Yingqi Gao and Yifu Liu and Xiaoxia Li and Xiaorong Shi and Yin Zhu and Yiming Wang and Shiqi Li and Wei Li and Yuntao Hong and Zhiling Luo and Jinyang Gao and Liyu Mou and Yu Li},
      year={2024},
      journal={arXiv preprint arXiv:2411.08599},
      url={https://arxiv.org/abs/2411.08599},
      primaryClass={cs.AI}
}

xiyan_mcp_server