I built DBeast: a PostgreSQL MCP server that gives AI assistants DBA-level tools

SN — Tue, 02 Jun 2026 21:25:33 +0000

AI assistants can write SQL.

But writing SQL is only a small part of working with a real database.

Most database work is investigation:

What tables exist?
How are they related?
Why is this query slow?
Which indexes are unhealthy?
Is autovacuum keeping up?
Are there risky permissions?
Which columns may contain sensitive data?
What would happen if we changed or deleted this data?

That is what I built DBeast for.

DBeast is a PostgreSQL MCP server that gives AI assistants structured DBA-style tools for understanding, analyzing, and troubleshooting PostgreSQL databases.

GitHub: https://github.com/snss10/DBeast

Why I built it

A lot of AI database integrations start with a simple tool:

execute_sql(query)

That is useful, but it is also incomplete.

A good database assistant should not just run SQL.

It should help you reason about the database.

It should inspect schemas, explain query plans, identify maintenance issues, detect security risks, analyze data quality, and help you understand the impact of changes before you make them.

That is the idea behind DBeast.

What DBeast does

DBeast exposes 21 MCP tools for PostgreSQL across areas like:

Schema discovery
ERD generation
Safe SELECT query execution
Query analysis
Query optimization hints
Write-impact previews
Database health checks
Maintenance analysis
Index and vacuum insights
Security audits
Sensitive data detection
Data quality reports
Duplicate detection
Replication monitoring
PostgreSQL configuration review
Audit logging

Instead of giving the assistant one giant database tool, DBeast gives it focused tools that map to real DBA workflows.

The bigger idea

I do not want AI assistants to be just SQL autocomplete.

I want them to become useful database copilots.

With DBeast, you can ask things like:

Show me how the sales schema is structured and generate an ERD.

Investigate why this dashboard query is slow and suggest indexes.

Review the public schema for maintenance issues, security risks, and data quality problems.

Compare table bloat, dead tuples, and index health across all schemas.

Find columns that may contain sensitive customer data.

Check replication lag and tell me if anything looks unhealthy.

That is much more useful than just:

Run this SQL.

Safety still matters

DBeast is designed to be safe by default.

SELECT queries can run with automatic row limits.

But write operations like:

INSERT
UPDATE
DELETE
DROP
TRUNCATE

are not executed directly.

They are analyzed as impact previews.

So instead of blindly running a risky query, your assistant can tell you:

what tables may be affected
how many rows may change
whether related objects are involved
what the risk level looks like
what you should review first

The human stays in control.

How it works

DBeast runs as a local MCP server.

AI assistant  ->  DBeast MCP server  ->  PostgreSQL

Your assistant talks to DBeast through the Model Context Protocol. DBeast talks to PostgreSQL using asyncpg.

It works with:

Local PostgreSQL
Docker PostgreSQL
AWS RDS / Aurora
Supabase
Neon
Railway
Render
Fly.io
SSH tunnels
AWS Secrets Manager

Quick start

git clone https://github.com/snss10/DBeast.git
cd DBeast
pip install -e .

Example MCP config:

{
  "mcpServers": {
    "dbeast": {
      "type": "stdio",
      "command": "python",
      "args": ["/absolute/path/to/DBeast/src/server.py"],
      "env": {
        "DATABASE_URL": "postgresql://user:password@localhost:5432/mydb"
      }
    }
  }
}

Then ask your assistant about your database.

Who this is for

DBeast may be useful if you:

use PostgreSQL
use Cursor, Claude Desktop, Windsurf, or another MCP client
want AI help understanding a database
want query and performance analysis
want safer database workflows
want a local tool instead of a hosted database agent

What I’m looking for

This is an early release, and I would love feedback from:

PostgreSQL users
DBAs
backend engineers
MCP builders
people experimenting with AI developer tools

I’m especially interested in:

what database workflows you would want next
whether setup is clear
what safety checks would make this more trustworthy
which MCP clients you want first-class examples for