PlanetScale’s MCP server exposes the parts of the platform that are awkward to hold in your head: which branches exist, how their schemas differ, and which queries are quietly costing you the most.
https://mcp.pscale.dev/mcp/planetscale
Claude Sonnet 4.5
Chat with 60+ AI models on the same workflow — switch to a different model mid-conversation and re-run the same prompt, or use Compare mode to put several side-by-side and balance quality vs. cost.
A PlanetScale service token or OAuth session. Scope it to the databases the agent should reach — tokens are per-organisation.
How models use it and what it is built for.
The server covers PlanetScale’s control surface alongside its data. An agent can list organisations, databases and branches, then read the schema of any branch — which matters on a platform where branching is the normal workflow and "what does the schema look like" is always a question about a specific branch. Query insights are readable, so a model can surface the statements with the worst latency or highest row-read counts and reason about which index is missing. Deploy requests, the mechanism by which schema changes reach production, are inspectable too. The combination suits an agent well: ask which queries regressed, read the schema they touch, and get a concrete index proposal with the reasoning attached, rather than a generic suggestion to add an index somewhere.
Typical tools an AI model can call. Exact names vary by version.
Copy any of these into MCP Agent Studio after connecting.
Which queries on the main branch have the worst p99 latency this week?
How does the schema on the add-billing branch differ from main?
Suggest an index for the slowest query, and explain the trade-off on write cost.
What deploy requests are open, and what tables do they touch?
This is not a single-model product: you get the same MCP connection with 60+ models (Claude, GPT, Gemini, DeepSeek, open-weight, and more), you can switch mid-conversation, and you can open Compare mode to run the same prompt against multiple models at once. The card above is a suggested starting point for this server — not the only choice.
Default pick for PlanetScale
Claude Sonnet 4.5
Index advice needs the schema, the query and the access pattern held together. Sonnet 4.5 reasons about composite column order; smaller models suggest single-column indexes reflexively.
Listing tools proves the server is reachable, not that a model can work with it. Evals go further: they read every tool on the server, write a test suite from its real schemas, and run it — code decides pass/fail on the responses (schema conformance, error codes, pagination, result caps) while a scoring model grades plain-English tasks driven through the tools.
Get a pass/fail report per tool with the evidence behind each verdict — and replay the same suite after every schema change. Destructive tools are excluded from the run.
Run evalsOpen MCP Agent Studio with the connection pre-filled. Add your token, pick any of 60+ models, and start chatting — no install required.
Open Agent StudioCommon questions about connecting, scoping and using it safely.
A hosted MCP server for PlanetScale’s serverless MySQL platform. It exposes databases, branches, schemas, query insights and deploy requests as tools, so an agent can explore and reason about your fleet.
Because on PlanetScale the schema is always a property of a branch, not of the database. An agent that does not name the branch is answering about an unspecified version of your schema — worth being explicit about in your prompts.
That depends entirely on the token you give it. Scope the service token to read-only, or to a development branch, and the question resolves itself. Never hand an agent a production-write token for exploratory work.
The Postgres server is a generic database connection you point at any instance. This one is platform-aware: it understands branches, deploy requests and PlanetScale’s own query insights, which are exactly the things a generic SQL connection cannot see.
It is one of the better fits for an agent, because the raw data is tedious and the pattern-matching is mechanical. Pulling the top twenty statements and correlating them against schema is a job models do well and humans postpone.