Feedback wanted: MCPg — safe PostgreSQL MCP server for AI agents

[devopam]

Hi Fellow Builders,

I built MCPg, an open-source PostgreSQL MCP server designed for AI agents and database workflows. It focuses on safe-by-default operation, PostgreSQL-aware tooling, and practical usability for Claude, Cursor, and other MCP clients.mcpservers+1

I’m looking for feedback on:

• Whether the tool surface feels useful or too broad.

• Whether the safety defaults are strong enough for real-world use.

• Whether the naming, docs, and examples make it easy for new users to start.

• What would make you trust it enough to use regularly in production or near-production setups.

What MCPg currently emphasizes:

• Read-only first behavior and guarded execution paths.

• SQL validation and safety checks.

• Support for both stdio and HTTP transports.

• Observability and operational tools for PostgreSQL workflows.

What I’d especially value feedback on:

1. The first-run experience.

2. Tool naming and discoverability.

3. Postgres-specific correctness.

4. MCP client compatibility.

5. What would make this more valuable to teams, not just individual users.

If you try it, please share:

• Which MCP client you used.

• What worked smoothly.

• What felt confusing.

• Any missing PostgreSQL operations or safety features you would expect.