Skip to main content
For an AI application that consumes MCP tools, start with MCP. This guide covers an AI application using the Agent API through its own backend. Your code controls which operations the model can request and validates their inputs before sending HTTP requests.

​
Connect your backend

Run the Agent API quickstart, then connect your application. Keep credentials, JWTs, and signing-key passwords in your backend. Give the model only the data needed for its task.

​
Define application actions

Expose a narrow action for each supported task, such as retrieving selected account details or inspecting a known token. Validate identifiers and authorization in code. A model-generated request is not proof that a user authorized it. Show the exact contract and role for review before requesting a signature. Obtain explicit user approval for payments or destructive actions as required by your application. Persist object IDs and verify completion.

​
Use the same guides and reference

Follow Build for the underlying workflow and the API reference for exact operations. Use documentation tools to help a coding agent find these instructions.