MCP SERVER + CI BOT
Production context while your agent codes. A CI backstop before it merges.
DBGorilla's MCP server lets your coding agent see your production database while it writes: schema, indexes, query stats, live contention, and how they change. The DBGorilla CI bot sees all of that plus your code, and reviews everything in the pull request that touches the database. The shape and behaviour of your database, never your rows.
IN THE EDITOR · AGENT PROPOSED
CREATE INDEX idx_events_created_at ON events (created_at);
MCP SERVER REPLIED
CREATE INDEX CONCURRENTLY idx_events_tenant_created
ON events (tenant_id, created_at DESC);Every read of this table filters on tenant_id first. An index on created_at alone scans 41.2M rows and discards 99.7% of them.
ON THE PULL REQUEST · CI BOT CHECKED
Index matches production access patterns. CONCURRENTLY: no lock on events during the build. Nothing else in this diff touches the database.
Same source of truth at both ends. In CI it is a fresh review: an objective judge whose only input from the developer is the code.
What each half actually sees.
The MCP server, while your agent codes
The schema of every table, and how it has drifted since
The queries that actually run, with their production stats
Live sessions, lock contention, infrastructure metrics
EXPLAIN against production, plan only, never executed
Experiments on an ephemeral clone with production statistics
The CI bot, before it merges
Everything the MCP server sees. It is the same server.
Plus the pull request diff, line by line
Plus the rest of your codebase, beyond the diff
DDL and generated SQL are verified on a clone before it comments
The lock a migration takes, at the table's live size
Three commands and a connected database.
Collector runs in your network. Docker, Kubernetes, RDS/Aurora, Azure, GCP or native Postgres.
What you'll see in the first ten minutes.
Your slowest queries, ranked.
Read from your real pg_stat_statements and query plans, not a synthetic benchmark.
What changed, and why.
Plan regressions, lock risk on migrations, bloat and stale statistics: each with the cause, in English.
The fix, tested on a clone.
Every recommendation is validated against a production-shaped copy before it reaches you.
“I just ran a benchmarked query with the indexes applied and 🤯 157x faster.”
Your first 30 days are Pro.
- 2,000 Gorilla Credits / month
- Up to 10 hosts · 100 DBs / cluster
- Full agent capabilities
- 30-day retention
- Regression detection & alerting
- Index impact simulator
- PR-CI bot & migration safety
- Isolated DB clones to test changes safely
✓When the 30 days end, the free tier keeps working, free forever: one cluster, the EXPLAIN analyzer, schema linter, index suggestions, and the MCP server.
Point it at one database and see.