# MCP API Tool Demo

Small demo showing how to expose an existing business API or internal data source as MCP tools.

This is intentionally simple and safe:

- no real customer data
- no external credentials
- local JSON data only
- two read-only tools

## Use Case

A client has order/customer data in an API, spreadsheet, or internal system, and wants an AI assistant to answer questions or call simple tools against that data.

This demo simulates that setup with local JSON files.

## Tools

- `get_order(order_id)`  
  Return order status, customer name, total and last update.

- `customer_summary(customer_id)`  
  Return a compact summary of a customer and recent order ids.

## Run

```bash
python -m venv .venv
source .venv/bin/activate
pip install -e .
python server.py
```

## Test

```bash
pip install -e ".[dev]"
pytest -q
```

Expected result:

```text
4 passed
```

## Quick Demo Without An MCP Client

```bash
python examples/run_demo.py
```

This prints sample outputs for:

- a known order
- a missing order
- a customer summary

## Example Questions A Client Could Ask

- What is the status of order `ord_5002`?
- Show me a short summary for customer `cust_1002`.
- Which recent orders need review?

## Notes For Real Client Work

For a real project, the JSON adapter would be replaced with:

- a REST API client
- a database query layer
- a Google Sheet / Airtable connector
- a private internal service

The MCP tool surface should stay small at first: one or two tools, clear input schema, read-only by default, and a short runbook for setup.

See `SECURITY_NOTES.md` for the security assumptions used in this demo.