Databases MCP Server (Access and SQLite 3)

A simple MCP server to let AI interact with Microsoft Access and SQLite 3 databases. Supports import/export with CSV and Excel files, and store human-readable notes about files.

WARNING: This server has full access to databases, so it can read and modify any data in it. Use with caution to avoid data loss!

Available Tools

Database management:

• list: List all active databases available in the server.
• create: Create a new database file (for Microsoft Access, copies the empty.mdb template).
• connect: Connect to an existing database file, or creates an in-memory database if the file is not specified.
• disconnect: Close a database connection. For in-memory databases, this will clear all its data.

Note: use connect tool with readOnly=true to open a database in read-only mode. For SQLite in-memory databases, read-only mode is not supported.

Data management:

• query: Execute a SQL query to retrieve data from a database.
• update: Execute a SQL query to insert/update/delete data in a database.
• import_csv: Imports data from a CSV file into a database table.
• export_csv: Exports data from a database table to a CSV file.
• import_excel: Imports data from an Excel file into a database table.

Notes management:

• read_notes: Reads notes from the specified file, or discovers notes in the specified directory.
• write_notes: Writes notes to the specified file, or linked to the specified database.

Note: Excel export is not implemented, use haris-musa/excel-mcp-server instead. The main problem is tracking the index of the rows and columns in the Excel file, to correctly import/export data to the same cells, and/or insert new rows/columns. In addition, merged cells complicate the process, it would be too complex to implement.

Supported Database Types

By default, database driver detection is based on file extension:

• Microsoft Access: .mdb and .accdb
• SQLite 3: .db, .sqlite, and .sqlite3

If database path is not specified (empty string), an in-memory SQLite database will be created.

To let the server open or create a database file without relying on the extension (for example, if your database file has a non-standard extension), you can specify the driver explicitly using the driver parameter in the connect or create tool.

• driver="access" for Microsoft Access
• driver="sqlite" for SQLite 3
• driver="auto" (default) automatically detects the driver based on the file extension

Project structure

Main files:

• server.py: MCP server implementation.

Tests:

• test_tools.py: Functions to test individual MCP tools.
• test_mcp.py: Tests all MCP tools in a typical workflow.

Documentation:

• README.md: This file, with general information about the project.
• LICENSE: MIT license.

Scouting scripts, used in the first stages to develop basic functionality:

• scouting_mdb.py: SQLAlchemy and pandas to interact with Microsoft Access databases.
• scouting_csv.py: SQLAlchemy and pandas to interact with CSV files.

TODO

• Add tool to create a new database, copying empty.mdb to the specified path.
• Add the ability to connect to multiple databases at the same time.
• Add tool to list all tables in the database.
• Add tools to import/export data from/to CSV files.
• Add tools to import data from/to Excel files.
• Add prompt to guide AI asking info to the user about the database.
• Store info about files (.AInotes files), to retrieve it later.
• Add driver parameter to manage databases without relying on file extension.
• Add tool to remember imported/exported CSV and Excel files.

Configuration

To use this MCP server with Claude Desktop (or any other MCP host), clone the repo and add the following to your config.json:

{
  "mcpServers": {
    "access-mdb": {
      "command": "uv",
      "args": [
        "run",
        "--with", "fastmcp",
        "--with", "pandas",
        "--with", "sqlalchemy-access",
        "--with", "openpyxl",
        "fastmcp", "run",
        "path/to/repo/server.py"
      ],
    }
  }
}

Dev note: to use with uvx, we need to create a package and publish it to PyPI.