Docs/AI Assistants (MCP)

From a plain-English question to ready-to-run SQL

Ask a question in plain English — the AI finds the right database, writes the SQL, runs it read-only, and hands you back both the answer and the query. You don't even need to remember your own schema.

  • Model: Claude Haiku 4.5 — the cheapest, fastest Claude.
  • Sample: sakila — the MySQL DVD-rental demo (films, customers, rentals, payments).

Not Opus. Not the top-tier Fable 5 (the Mythos-class model above Opus). Every join below ran on the budget model — that's the point.

Round 1 — ask in plain English

You: I have a database about movie rentals somewhere in my MySQL connections — I don't remember which one. Find it and show me the top 5 films by total revenue.

You named nothing — no connection, no database, no table. The AI finds it by chaining read-only tools:

list_connections → several MySQL connections
list_databases   → spots sakila
list_tables      → film, inventory, rental, payment

Then it writes the four-table join, runs it read-only (30 ms), and answers:

AI: Top 5 films by revenue in sakila:

FilmRevenueRentals
TELEGRAPH VOYAGE$231.7327
WIFE TURN$223.6931
ZORRO ARK$214.6931
GOODFELLAS SALUTE$209.6931
SATURDAY LAMBS$204.7228

Another question, different tables, same flow:

You: And who's my single biggest-spending customer?

AI: KARL SEAL — $221.55 across 45 payments (runner-up: Eleanor Hunt, $216.54). Joined customer → payment.

Round 2 — keep the queries

You: Give me those queries so I can run them myself in the SQL Console.

AI: Top films by revenue:

SELECT f.title,
       ROUND(SUM(p.amount), 2)      AS revenue,
       COUNT(DISTINCT r.rental_id)  AS rentals
FROM film f
JOIN inventory i ON i.film_id = f.film_id
JOIN rental    r ON r.inventory_id = i.inventory_id
JOIN payment   p ON p.rental_id = r.rental_id
GROUP BY f.film_id, f.title
ORDER BY revenue DESC
LIMIT 5;

And top customers by spend:

SELECT c.first_name, c.last_name,
       ROUND(SUM(p.amount), 2) AS total_spent,
       COUNT(*)                AS payments
FROM customer c
JOIN payment p ON p.customer_id = c.customer_id
GROUP BY c.customer_id, c.first_name, c.last_name
ORDER BY total_spent DESC
LIMIT 5;

Paste either into the SQL Console — they already returned real rows above.

What this shows

  • You ask in plain language. The AI walks connection → database → tables itself — no schema lookup, no recalling table names.
  • The number is real. It runs the query under MCP's read-only guard before answering, against your actual data.
  • You keep the query. Handed back ready to paste and reuse — the AI never touches your write path.

What's next

PreviousSafety & PrivacyNextExample: federated validation