Review your AI agent's database changes on your schema, with plain-language risk, before they run.
Scored across 6 tools
Each tool has a distinct read/write target: project listing, schema read, issue listing, proposal listing, single-proposal read, and proposal creation. The only pair with adjacent purpose is get_proposal vs list_proposals, but the singular/plural distinction and the descriptions (status + migration SQL vs recent proposals) make them separable, and propose_change being the sole write tool is clear.
All six tools follow a strict verb_noun snake_case pattern: get_proposal, get_schema, list_issues, list_projects, list_proposals, propose_change. The verb matches the operation type consistently (list for collections, get for singular resources, propose for the mutating/review action).
Six tools is tightly scoped to the domain: three list/get reads over projects and proposals, one schema read, one diagnostics read, one write. Nothing looks redundant or bolted on, and the surface is small enough to fit comfortably in an agent's working set.
The core lifecycle (inspect project, read schema, propose change, review proposal, read approved SQL, diagnose issues) is covered end to end. Minor gaps exist: there is no way to create or rename a project, no way to withdraw or reject a proposal, and no migration-history/apply confirmation, though the human-in-the-loop design intentionally leaves migration application to the project repo.
Returns a proposal’s status. When approved, also returns the migration SQL to apply (up) and to undo it (down) for the chosen engine. Write the up SQL into the project’s migrations exactly as given.
Inputs: proposal_id (required), dialect.
Returns the current schema of a project as SQL for its database engine (default) or as JillionDB JSON.
Inputs: project_id (required), format, dialect.
Lists design problems in the current schema: missing keys, unindexed relationships, type mismatches and engine limits.
Inputs: project_id (required), dialect.
Lists the JillionDB projects this key can see, with their database engine and table count.
Inputs: no named parameters.
Lists recent proposals for a project, newest first.
Inputs: project_id (required), status.
Proposes a schema change for the user to review in JillionDB. Nothing changes until they approve it. Send exactly one of "operations" (preferred: renames are explicit, so no data is lost) or "sql" (the complete desired schema as CREATE TABLE statements; a dropped and re-added column is flagged as a possible rename). Returns the risk of each change and a review link to give the user.
Each operation is an object with "op" and the fields below. Names are table and column names as they appear in get_schema.
create_table: table, columns [{name, type, nullable?, default?, unique?, primary_key?, comment?, references?}], primary_key? [names], comment?
drop_table: table
rename_table: table, to
add_column: table, column {name, type, nullable?, default?, unique?, comment?, references?}
drop_column: table, column
rename_column: table, column, to
alter_column: table, column, type?, nullable?, default?, unique?, comment?
add_index: table, columns [names], unique?, name?
drop_index: table, name or columns
add_foreign_key: table, columns [names], references {table, columns? (defaults to its primary key), on_delete?}
drop_foreign_key: table, columns or name
set_primary_key: table, columns
add_check: table, expression, name?
drop_check: table, name or expression
set_comment: table, column?, comment Types: uuid, text, varchar(n), char(n), smallint, integer, bigint, serial, bigserial, decimal(p,s), float, double, boolean, date, time, timestamp, timestamptz, json, binary, enum('a','b'). Defaults: a value (string, number, true/false), "now", "uuid", "autoincrement", {"value": ...} for literal text that looks like a keyword, or {"sql": "..."} for a raw expression. references: {table, column?, on_delete?: cascade | restrict | set null | set default | no action}.
Inputs: summary (required), project_id (required), sql, operations.
Recent tool additions, removals, and schema changes observed during successful MCP inspections.
Loading tool schema changelog…
Deterministic safety, correctness & cost gate that vets Postgres SQL before your AI agent runs it.
AI agents propose database changes as reviewable requests — no direct write access.
Database guardrails for AI agents: check SQL against your Vericto rules before it runs.
AI agents need permission before production SQL writes. Pilot $100 · Gateway $299. Lint≠authorize.
Add one secure layer between your agents and this server.
