Stop tuning queries manually.
Let AI iterate until it's fast.
An MCP server that autonomously rewrites, indexes, and benchmarks your SQL until it can't get any faster.
Your Favorite AI Agent
An MCP server that autonomously rewrites, indexes, and benchmarks your SQL until it can't get any faster.
Click the sidebar icon on the top left corner
Click your profile → Settings → Connectors
Click "Add custom connector" and enter Name: ExoBench
https://connect.exobench.ai/mcpThen click Add to save the connector.
Open the sidebar again and click New Chat.
For example:
This query is slow. Can you use ExoBench to make me an optimized version or a good index?
SELECT customer_id, SUM(amount) FROM orders
WHERE status = 'pending' GROUP BY customer_id;
Scroll to the bottom of the chat and click Connect.
Click "Continue with Github" → "Authorize ExoQuery MCP" → "Always Allow"
Don't see your AI tool? Any MCP Client →
ExoBench uses the Model Context Protocol (MCP) to give your AI tool direct access to a benchmarking database. The AI drives the optimization loop autonomously.
Give your AI any SQL query you want optimized.
Rewrites, creates indexes, benchmarks, and repeats until convergence.
Receive the optimized query, indexes, and a full performance report.
Finds missing indexes you didn't know you needed. Analyzes EXPLAIN plans and creates optimal single and composite indexes.
Restructures joins, subqueries, and filter ordering for optimal execution plans. Not cosmetic changes, structural ones.
Proposes covering indexes and denormalization when the query can't be optimized further with rewrites alone.
Doesn't stop at the first improvement. Loops until the benchmarker confirms no further gains are possible.
| Manual Optimization | AI Chatbot |  ExoBench | |
|---|---|---|---|
| Runs real EXPLAIN ANALYZE | You do it yourself | No, guesses from syntax | Yes, on real Postgres |
| Tests at multiple data scales | Prohibitively slow | Can't execute anything | Automatic, one request |
| Tests data distribution impact | Use the skew you have | Can't test, only speculate | Specify skew, see real plans |
| Verifies suggestions work | If you benchmark manually | No way to verify | Every suggestion benchmarked |
| Iterates on optimizations | You iterate manually | Copy-paste loop | AI iterates autonomously |
| Time per query | Hours to days | 30-60 min (unverified) | Minutes (verified) |
| Your data stays private | You run on staging | You paste schema into chat | Ephemeral DBs, synthetic data |
ExoBench works with the databases you already use.
Optimize this query. It's a department salary report that runs nightly and is getting slow. Our employees table has 245k rows with hire dates going back to 2015. The nightly job filters for the last ~18 months of hires. Hire dates are spread over 10 years. SELECT d.name AS department, COUNT(*), AVG(e.salary) FROM employees e JOIN departments d ON d.id = e.dept_id WHERE e.hire_date > '2023-06-01' GROUP BY d.name
Added covering index on (hire_date, dept_id) INCLUDE (salary) to enable Index Only Scan — zero heap fetches
"I built ExoBench because I managed thousands of SQL queries in the financial services industry and got tired of hand-tuning every one of them."
ExoBench is currently in invite-only preview. Request access to get started.
Applications reviewed within 24–48 hours.