Skip to main content
The PlanetScale MCP server gives AI agents access to production query metrics (Insights) and Schema Recommendations alongside your application code. This combination enables a self-improving optimization loop: an agent identifies the highest-impact performance issues, finds the relevant code, makes improvements, and opens a pull request. This works for both Postgres and Vitess databases.

​
How it works

  1. The agent connects to PlanetScale through the MCP server and pulls Insights data or open Schema Recommendations.
  2. Identifies the highest-impact queries or schema issues.
  3. Searches your codebase to find where those queries originate.
  4. Makes the change in the codebase.
  5. Benchmarks the change in a development environment to validate the improvement.
  6. Opens a pull request with the fix and an explanation of the expected impact.
This can run as a one-off task or on a recurring schedule. Cursor Automations are a great way to run this daily — they run cloud agents in the background on a schedule and support MCP server connections, making them ideal for this workflow.

​
Prerequisites

These workflows only need Insights and Schema Recommendations access. You can use the insights-only MCP server (https://mcp.pscale.dev/mcp/planetscale-insights-only) to avoid exposing query execution tools to the agent.

​
Getting started

For best results, take an iterative approach:
  1. Run the prompt manually first. Use one of the example prompts below in your AI coding tool and review the pull request it produces.
  2. Refine with guardrails. Based on the results, add rules specific to your codebase — which directories to touch, which ORM patterns to follow, linter and test requirements, migration commands, etc.
  3. Repeat a few times. Run it again with your refined prompt. Nudge the agent where needed. Each iteration helps you build a prompt that consistently produces high-quality PRs.
  4. Graduate to a schedule. Once the agent reliably produces good results, set it up as a Cursor Automation on a daily or weekly schedule.
Always review pull requests before merging. Automated agents can surface valuable optimizations, but a human should verify that the changes are safe for your workload.

​
Example prompt: Insights optimization

This prompt instructs the agent to review your Insights data, find the highest-impact queries, and optimize them. Customize the rules section for your framework, ORM, and linter.
Adapt the rules to your stack. For example, a Rails project might add: “Run migrations with rails db:migrate to generate the schema.rb file” and “Run rubocop and fix any lint errors.”

​
Example prompt: Schema Recommendations

This prompt instructs the agent to evaluate open Schema Recommendations and implement them. Customize the rules section for your framework and workflow.

​
Tips for effective automations

  • One fix at a time. Each run should address a single query or recommendation. This keeps pull requests small, reviewable, and easy to revert.
  • Benchmark in development. Instruct the agent to run EXPLAIN or time queries on a development branch before and after the change. This validates the optimization and gives reviewers concrete data.
  • Use AGENTS.md to provide context. Include your PlanetScale org, database, and branch so the agent doesn’t have to scan through all your organizations.
  • Adapt prompts to your stack. Add rules for your specific framework, ORM, migration tool, and linter. The more specific the guardrails, the better the output.
  • Review every PR. Automated agents surface real improvements, but schema and query changes require human judgment before hitting production.

​
Need help?

Get help from the PlanetScale Support team, or join our Discord community to see how others are using PlanetScale.