Summary
Read-only is the tool's own contract, not something the prompt has to keep reminding the model about.
There is one tool and it runs read-only SQL, so nothing an agent composes can write, and a badly formed statement costs a failed query rather than a changed row. Which schemas are visible belongs to the server rather than to the query, which means the boundary of what can be seen is decided once, by whoever started it.
What it is
A single-tool PostgreSQL bridge that executes read-only SQL over the schemas the server was started against.
What you get
- query, which runs a read-only SQL statement and returns the result.
- More than one schema visible to the same statement, so a join across them is one call.
Requirements
Nothing to supply — no account, no key. A reachable PostgreSQL database and the list of schemas to expose, both settled when the server starts rather than per call.
Setup effort
One command — npx -y mcp-server-postgres-multi-schema <database-url> [schemas]