UnderDoc Tutorial - Expense Analytics using GenAI and MCP server for SQLite DB

What is UnderDoc Tutorial - Expense Analytics using GenAI and MCP server for SQLite DB?

This tutorial demonstrates how to extract structured expense data from receipt, invoice, and demand-note images using UnderDoc, store it in a local SQLite database, and then perform natural‑language expense analytics by chatting with Claude Desktop via an MCP server for SQLite. It is intended for developers and users who want to leverage GenAI and the Model Context Protocol to query financial data.

How to use UnderDoc Tutorial - Expense Analytics using GenAI and MCP server for SQLite DB?

First, obtain an UnderDoc API key, install uv, Python 3.12, and the SQLite CLI. Clone the tutorial repository, set up a Python virtual environment, and place expense images in the receipt-images folder. Run python main.py to extract data and save it to metabase-data/underdoc.db. Then install Claude Desktop, clone the reference MCP server for SQLite from the modelcontextprotocol/servers repository, and configure it in Claude Desktop to query the database. Finally, ask natural‑language questions about your expenses.

Key features of UnderDoc Tutorial - Expense Analytics using GenAI and MCP server for SQLite DB

• Extracts structured data (shop name, amount, currency, category) from images.
• Stores extracted data in a local SQLite database.
• Uses Claude Desktop as the GenAI interface for analytics.
• Integrates the open‑source MCP server for SQLite.
• Supports multi‑language receipt/invoice images.
• Provides sample images and a pre‑populated database for testing.

Use cases of UnderDoc Tutorial - Expense Analytics using GenAI and MCP server for SQLite DB

• Automating expense data entry from photographed receipts.
• Analyzing spending patterns by category, vendor, or date using plain English.
• Learning how to combine a document‑understanding API with an MCP server for database queries.
• Building a local, no‑code expense analytics dashboard via conversational AI.

FAQ from UnderDoc Tutorial - Expense Analytics using GenAI and MCP server for SQLite DB

What prerequisites are needed to run this tutorial?

You need a MacOS laptop or desktop, uv (Python runtime/package manager), git, Python 3.12, the SQLite command‑line interface, and Claude Desktop. An UnderDoc API key is also required; you can get a free tier key by signing up at the UnderDoc developer portal.

How is the expense data stored and where does it live?

The extracted data is persisted in a local SQLite database file (underdoc.db) located inside the metabase-data folder of the cloned repository. No data is sent to any external server beyond the UnderDoc API call.

Which MCP server is used for SQLite queries?

The tutorial uses the official MCP reference server for SQLite from the modelcontextprotocol/servers GitHub repository. It is run via uv and configured in Claude Desktop to enable natural‑language interaction with the database.

What are the runtime requirements for the MCP server?

The SQLite MCP server requires uv and Python 3.12. Claude Desktop must be installed on macOS. The tutorial was tested on a Mac, and the server communicates locally over standard I/O with the GenAI client.

Can I use my own expense images instead of the samples?

Yes. Place your own receipt, invoice, or demand‑note images into the receipt-images folder before running the extraction script. The script processes all images in that folder.