Docs

Connect Veto to your agent

Veto is an MCP server. Point your AI coding agent at the endpoint, and it can ask for a deterministic verdict before it runs any SQL. No SDK, no database connection, no API key required for the free tier.

MCP endpoint · Streamable HTTP · https://vetosql.com/mcp

1. Add the MCP server

Veto speaks the Model Context Protocol over Streamable HTTP. Add it to any MCP-capable client (Claude Code, Cursor, Cline, …) with a single entry:

{
  "mcpServers": {
    "veto": { "url": "https://vetosql.com/mcp" }
  }
}

In Claude Code you can also run claude mcp add --transport http veto https://vetosql.com/mcp. In Cursor, add the same JSON under Settings → MCP. That's the whole setup for the free tier.

2. The `analyze_sql` tool

Your agent calls analyze_sql before executing a statement. Inputs:

Field	Required	Description
`sql`	yes	The Postgres SQL or migration to vet.
`schema`	no	DDL for the relevant tables. Supplying it enables `EXPLAIN`-based cost analysis on a disposable scratch database.
`rowCountHints`	no	Approximate row counts per table, to sharpen cost estimates.

It returns a structured result: an overall verdict plus a list of findings, each with a stable dotted id (e.g. destructive.delete_without_where) your pipeline can branch on. The same SQL always yields the same verdict — there is no LLM in the loop.

3. Verdicts

Verdict	Meaning
`ok`	No rule fired. (Not a guarantee of safety — see the terms.)
`warn`	Something worth a human glance: `SELECT *`, a lock-heavy `ALTER`, a correctness trap, a large sequential scan.
`block`	Genuine data-loss risk: unscoped `DELETE`/`UPDATE`, `DROP TABLE`, `TRUNCATE` — including ones hidden inside CTEs.

4. Pro: authentication

Pro raises your rate limit and unlocks custom org policies. Authenticate by sending your key as a bearer token on the MCP endpoint:

{
  "mcpServers": {
    "veto": {
      "url": "https://vetosql.com/mcp",
      "headers": { "Authorization": "Bearer YOUR_PRO_KEY" }
    }
  }
}

5. Pro: custom org policies

The built-in rules don't know that payments is sacred in your database. Custom org policies let your team add per-table rules that Veto enforces on top of the built-ins. Your agent can author the JSON from a plain-language instruction and call the set_policies tool with your Pro key:

[
  { "table": "payments", "operations": ["delete", "truncate"], "action": "block",
    "message": "Never delete from payments." },
  { "table": "audit_*", "operations": ["truncate"], "action": "block" }
]

Policies are stored on your key, so every later analyze_sql call made with it enforces them automatically — no per-call config. They are declarative data: validated, matched by glob (not regex), and never executed.

6. Good to know

Never connects to your database. Cost analysis runs on a throwaway scratch Postgres inside a transaction that's always rolled back.
Deterministic. Same input, same verdict — testable and auditable, with no model drift.
Postgres-native. Works with Supabase, Neon, RDS, or self-hosted, and any migration tool.

More background in the FAQ and the blog.

That's it — point your agent at https://vetosql.com/mcp and it can ask for a verdict before every query.

Connect Veto via MCP