Veto vs Boyce

Boyce (by boyce-io) is an MCP server that gives AI agents structured database intelligence — deterministic SQL analysis, NULL trap detection, and EXPLAIN pre-flight. Veto overlaps on correctness and cost, but optimizes for a different job: a hard ok / warn / block safety gate before execution.

At a glance

BoyceVeto
PositioningDatabase intelligence for agents (MIT)SQL safety & cost oracle — pre-execution gate
Deterministic analysisYes — NULL traps, EXPLAIN pre-flightYes — rules + optional EXPLAIN on scratch Postgres
Destructive opsVaries by implementationFirst-class: unscoped DELETE/UPDATE, DROP, TRUNCATE, CTE-hidden deletes → block
Verdict modelStructured intelligence / findingsExplicit ok / warn / block for pipeline gating
Custom org rulesPro: set_policies per-table rules on your key
DeploymentSelf-hosted MCP (typical)Hosted remote MCP at vetosql.com/mcp — no install

When Boyce fits

You want a MIT-licensed MCP server you run yourself, focused on giving agents rich, deterministic database context — schema intelligence, NULL semantics, query planning — as part of a broader agent toolkit.

When Veto fits

You need a gate, not just intelligence: the agent must get a block verdict on DELETE FROM payments before it runs, with stable finding ids for CI and audit. You want a hosted endpoint (free tier, no key) and optional Pro policies like "never DELETE from payments".

Can you use both?

Possibly — if Boyce supplies context and Veto is the last check before execution. They solve overlapping but not identical problems. Evaluate whether you need a general database brain, a safety gate, or both.

Point your agent at Veto before it touches the database.

Connect via MCP All comparisons