These are prompts to try once you have connected the SurrealDB MCP Server to your AI tool. Each one explains what the assistant does behind the prompt, so you know what to expect before you ask.
See what you have
What SurrealDB instances do I have, and which ones are paused?
The assistant lists your organisations, picks the one you meant or asks which, and reports every instance with its region, size, version, and state. A good first prompt, because it confirms the connection is working.
Which of my instances is on an old version of SurrealDB?
It reads the version of each instance and compares them against the versions available to deploy.
Deploy an instance
I need a small instance in Frankfurt for a side project. What would it cost, and can my organisation deploy it?
The assistant looks up the regions and instance types available to you, puts a configuration together, prices it, and checks whether the organisation can deploy at all: free instances left on the plan, billing details filled in, payment method on file. You get the cost and any blockers before anything is created.
Go ahead and deploy it, and tell me how to connect once it is up.
Only after you say so does it create the instance. It then reports the new instance and how to reach it. If billing is not set up yet, it asks for the contact details it needs and hands you a secure checkout link to add a card yourself.
Clients with a prompt menu also offer a Deploy a Cloud Instance wizard, which walks the same path step by step.
Query and shape your data
In my dev instance, create a
tasktable with a title and a done flag, add a couple of rows, then show me the ones that are not done.
The assistant selects the namespace and database, writes the SurrealQL, and shows you both the statements and the results. The instance has to be running and on SurrealDB 3.1 or later.
How many orders did we take last week, grouped by status?
It inspects the schema to find the right table and fields, then writes the aggregate query rather than asking you for the shape of your data.
Something looks wrong with the
customertable. What does its schema actually say?
It describes the table, its fields, and its indexes, which is often faster than opening a dashboard to check a definition.
Investigate a problem
My production instance felt slow this morning. Have a look and tell me what you find.
The assistant checks the instance's status, pulls metrics for the window you describe, and reads the logs around it. Because it can also query the database, it can follow a hunch about a missing index in the same conversation.
Compare CPU and memory over the last day against the same period last week.
It requests both windows and summarises the difference.
Manage your team
Invite alex@example.com to the Acme organisation as a developer, and show me who else has access.
It lists the roles the organisation offers, sends the invitation with the role you chose, and reports current members along with any invitations still outstanding.
Keep costs in check
What have we spent on SurrealDB this month, and which instance accounts for most of it?
The assistant reads usage for the organisation and for each instance, and pulls up recent invoices.
Pause anything in the sandbox organisation that nobody has touched this week.
It works out which instances are idle from their usage, tells you what it plans to pause, and waits for you to confirm. Paused instances keep their data.
Give an agent long-term memory
Set up a Spectron context for this project in Frankfurt, then remember that we settled on event sourcing for the orders service.
The assistant creates the Spectron context, registers a scope path for the project, and stores the decision in it.
What do we already know about the orders service?
It recalls what has been stored for that project and answers from it, so a decision made weeks ago still turns up in today's conversation.
Getting better results
Name the instance or organisation when you have more than one. Otherwise the assistant has to stop and ask.
Ask for a plan first on anything expensive or irreversible: "tell me what you would change, then wait".
Let it check before it deploys. Readiness and cost checks cost nothing, and they stop a deploy failing halfway.
Keep approvals on write tools. Auto-approving read-only tools saves clicks; auto-approving the rest removes the pause that lets you catch a mistake.
Next steps
SurrealDB MCP Server — what each group of tools does, and how signing in works
Agent Skills — teach your assistant idiomatic SurrealQL to go with these tools
Embedded MCP — the same data tools against a database you run yourself