BizDesk MCP

✅ Verified end-to-end on Windows with local Ollama (llama3.1, num_ctx=8192) — every command below is from a real run.

BizDesk MCP is a Model Context Protocol server that gives Claude (or any MCP client) structured, read-mostly access to a small-business back office: customers, invoices, and account notes. Ask Claude "who owes me money?" and it calls overdue_invoices() instead of guessing.

MCP is how AI assistants get real hands on real business data — and this repo is the pattern I use for client systems, shrunk to a clean, runnable example.

What it exposes

Design choices worth copying: tools return structured rows, not prose; the only write path (add_note) is append-only; and the server ships instructions so the model knows the house rules before it calls anything.

Related MCP server: Frihet MCP Server

Quickstart

pip install -r requirements.txt
python seed.py        # creates bizdesk.db (8 customers, ~50 invoices)

Claude Desktop — add to claude_desktop_config.json (see examples/):

{ "mcpServers": { "bizdesk": { "command": "python", "args": ["/abs/path/server.py"] } } }

Claude Code —

claude mcp add bizdesk -- python /abs/path/server.py

Restart the client, then try: "Which invoices are more than 30 days overdue, and draft a polite reminder for the worst one?" — watch it chain overdue_invoices → bizdesk://customers/... → a grounded draft.

Point it at real data

bizdesk.db is plain SQLite. Swap the queries in server.py for your Postgres/Airtable/QuickBooks layer and the MCP surface stays identical — the client never knows the backend changed. That's the point of the protocol.