Simple database assistant with Claude

Downloading a good skill or installing an MCP server? That’s what I imagined I’d need to let Claude connect to a database, then ask it to run any kind of SQL queries I wouldn’t have to write myself.

But actually, the only thing I had to do was to prompt: Connect to postgresql://myuser@localhost:5433/mydb. Claude will simply perform a SELECT version(); or \conninfo and confirm it can access the database and execute subsequent psql commands. Everything works thanks to my credentials being in ~/.pgpass.

Here’s an example of Claude inspecting a database structure using only psql commands:

❯ Inspect the `demo` schema and explain what it is. Display a schema.

[Claude inspects tables, columns, constraints...]

● What is the demo schema?      
                                                                                           
  It's a self-contained sample e-commerce dataset — a sandbox with 5 fictional users, 8 products, 200 orders, and 1000 analytics events. It has no dependencies on any other schema in the database. It's clearly
  meant for development, demos, or testing queries against a realistic-looking data model that uses PostgreSQL-specific types (jsonb, text[]).                                                                     
   
  Entity-Relationship Diagram                                                                                                                                                                                      
                  
  ┌──────────────────────────────┐
  │           users (5)          │
  ├──────────────────────────────┤
  │ PK id          serial        │
  │ UQ username    text NOT NULL  │
  │    email       text NOT NULL  │
  │    created_at  timestamptz   │
  │    metadata    jsonb          │
  │    ┌─────────────────────┐   │
  │    │ plan: free|premium   │   │
  │    │ country: FR|US|DE    │   │
  │    └─────────────────────┘   │
  └──────────┬──────────┬────────┘
             │          │
             │          │
       ┌─────┘          └──────────────────────┐
       │ FK user_id                   FK user_id│
       ▼                                        ▼
  ┌────────────────────────┐    ┌──────────────────────────────┐
  │     orders (200)       │    │        events (1000)         │
  ├────────────────────────┤    ├──────────────────────────────┤
  │ PK id        serial    │    │ PK id          serial        │
  │ FK user_id   → users   │    │ FK user_id     → users       │
  │    status    text       │    │    event_type  text NOT NULL  │
  │    total     numeric    │    │    payload     jsonb          │
  │    created_at timestamptz│   │    created_at  timestamptz   │
  ├────────────────────────┤    ├──────────────────────────────┤
  │ status values:         │    │ event_type values:           │
  │  pending | confirmed   │    │  login | page_view | search  │
  │  shipped | delivered   │    │  add_to_cart | checkout      │
  │  cancelled             │    └──────────────────────────────┘
  └───────────┬────────────┘
              │
              │ FK order_id
              ▼
  ┌──────────────────────────────┐
  │      order_items (600)       │
  ├──────────────────────────────┤
  │ PK id          serial        │
  │ FK order_id    → orders      │
  │ FK product_id  → products    │
  │    quantity    int NOT NULL   │
  │    unit_price  numeric        │
  └──────────────┬───────────────┘
                 │
                 │ FK product_id
                 ▼
  ┌──────────────────────────────┐
  │       products (8)           │
  ├──────────────────────────────┤
  │ PK id          serial        │
  │    name        text NOT NULL  │
  │    category    text NOT NULL  │
  │    price       numeric        │
  │    stock       int (def 0)   │
  │    tags        text[]         │
  │    created_at  timestamptz   │
  ├──────────────────────────────┤
  │ categories: electronics      │
  │             furniture        │
  └──────────────────────────────┘

  Indexes

  Only primary keys and the username unique constraint are indexed — no secondary indexes on foreign keys or query columns. This is consistent with a lightweight demo/dev dataset not optimized for production
  query patterns.

✻ Worked for 31s