MCP-FI

Query any database in plain English using Claude.

MCP-FI is a Model Context Protocol (MCP) server that connects Claude to your data sources — SQLite, CSV, PostgreSQL, and MySQL. Instead of writing SQL, you just ask questions and Claude figures out the rest.

---

How It Works

You: "Which customers have the highest lifetime value?"
        ↓
Claude calls summarise_data → understands your schema
        ↓
Claude calls query_data → generates and runs the SQL
        ↓
Claude explains the results in plain English

No SQL knowledge required.

---

Features

• Multiple data sources — SQLite, CSV, PostgreSQL, MySQL
• Three tools exposed to Claude:
  • summarise_data — understand table structure before querying
  • query_data — run SQL against any configured source
  • compare_data — diff two sources and find differences
• Plain English queries — Claude writes the SQL for you
• Local or deployed — run locally via Claude Desktop or deploy via SSE for others to use

---

Requirements

• Python 3.12+
• Claude Desktop (for local use)
• Node.js (for MCP Inspector testing)

---

Installation

1. Clone the repo:

git clone https://github.com/yourusername/mcp-fi.git
cd mcp-fi

2. Create and activate a virtual environment:

python -m venv venv
source venv/bin/activate        # Mac/Linux
venv\Scripts\activate           # Windows

3. Install dependencies:

pip install -r requirements.txt

4. Set up your config:

cp config.example.json config.json

Then edit config.json with your actual data source paths and credentials (see Configuration below).

---

Configuration

Edit config.json to define your data sources. Each source needs a unique name that you'll use when asking Claude questions.

{
  "sources": {
    "my_sqlite_db": {
      "type": "sqlite",
      "path": "/absolute/path/to/your/database.sqlite"
    },
    "my_csv_data": {
      "type": "csv",
      "data_dir": "/absolute/path/to/your/csv/folder/"
    },
    "my_postgres": {
      "type": "postgres",
      "dsn": "postgresql://user:password@localhost/dbname"
    },
    "my_mysql": {
      "type": "mysql",
      "host": "localhost",
      "port": 3306,
      "database": "mydb",
      "user": "root",
      "password": "yourpassword"
    }
  }
}

Important: Always use absolute paths — relative paths will not work when Claude Desktop launches the server.

You can define as many sources as you need. Mix and match types freely.

---

Connect to Claude Desktop

Open your Claude Desktop config file:

Mac:

open ~/Library/Application\ Support/Claude/claude_desktop_config.json

Windows:

%APPDATA%\Claude\claude_desktop_config.json

Add the mcpServers block:

{
  "mcpServers": {
    "mcp-fi": {
      "command": "/absolute/path/to/mcp-fi/venv/bin/python",
      "args": [
        "/absolute/path/to/mcp-fi/server.py"
      ],
      "cwd": "/absolute/path/to/mcp-fi"
    }
  }
}

Restart Claude Desktop. You should see a 🔨 hammer icon in the chat input — that confirms MCP-FI is connected.

---

Usage Examples

Once connected, just ask Claude questions in plain English:

Explore your data:

What tables are available in my_sqlite_db?

Query data:

Show me the top 10 customers by total orders from my_postgres

Filter and aggregate:

What is the average net worth by income level in my_csv_data?

Compare sources:

Compare the users table between my_sqlite_db and my_postgres — are they in sync?

Complex analysis:

Which products in my_mysql have been ordered more than 100 times 
and what is their average rating?

Claude will automatically call summarise_data to understand your schema, then query_data to fetch results, and finally explain everything back to you clearly.

---

Testing

Run the full test suite:

pytest tests/ -v

Run a specific connector test:

pytest tests/test_sqlite_connector.py -v
pytest tests/test_csv_connector.py -v
pytest tests/test_postgres_connector.py -v
pytest tests/test_mysql_connector.py -v

The suite covers 116 tests across all connectors, tools, router, and server — all using isolated temporary databases so no real data is touched.

---

Test With MCP Inspector

Before connecting to Claude Desktop, you can test all tools interactively:

npx @modelcontextprotocol/inspector python server.py

Open the URL shown in your terminal, click Connect, then Tools to see and run all three tools manually.

---

Deployment (SSE)

To deploy MCP-FI as a shared server others can connect to:

1. Run in SSE mode:

python server.py --sse --port 8000

2. Deploy to Railway:

Create a Procfile:

web: python server.py --sse --port $PORT

Push to GitHub, connect the repo to Railway, and deploy. You'll get a public URL like:

https://mcp-fi-production.up.railway.app

3. Others connect to your server by adding this to their claude_desktop_config.json:

{
  "mcpServers": {
    "mcp-fi": {
      "command": "npx",
      "args": [
        "-y",
        "mcp-remote",
        "https://mcp-fi-production.up.railway.app/sse"
      ]
    }
  }
}

---

Project Structure

mcp-fi/
├── connectors/
│   ├── base_connector.py      # Abstract contract all connectors implement
│   ├── sqlite_connector.py    # SQLite support
│   ├── csv_connector.py       # CSV support via DuckDB
│   ├── postgres_connector.py  # PostgreSQL support
│   ├── mysql_connector.py     # MySQL support
│   └── router.py              # Routes source names to correct connector
├── tools/
│   ├── query_tool.py          # Run SQL against any source
│   ├── summarise_tool.py      # Describe table structure
│   └── compare_tool.py        # Diff two sources
├── tests/                     # 116 tests across all layers
├── data/                      # Put your SQLite and CSV files here
├── server.py                  # Entry point — stdio and SSE transport
├── config.json                # Your data sources (gitignored)
├── config.example.json        # Template for config.json
└── requirements.txt           # Python dependencies

---

Adding a New Data Source

1. Add your source to config.json
2. Ask Claude: "What tables are in my_new_source?"

That's it. No code changes needed.

---

Supported Data Sources

Type	Config key	Notes
SQLite	"type": "sqlite"	Provide absolute path to .sqlite file
CSV	"type": "csv"	Provide absolute data_dir — each CSV becomes a table
PostgreSQL	"type": "postgres"	Provide dsn connection string
MySQL	"type": "mysql"	Provide host, port, database, user, password

---

License

...