Extension APIs¶
Extension APIs connect AI Studio processing capabilities to external systems or make external capabilities available to agents. This section covers three integration patterns:
Integration |
Direction |
Use case |
|---|---|---|
External application → AI Studio |
Run parsing, segmentation, embedding, information extraction, or a custom operator through a synchronous REST call |
|
AI Studio ↔ MCP service |
Bind tools from a remote MCP service to an agent; where supported by the deployment, expose an agent as an MCP server |
|
AI Studio → External receiver |
Deliver processing results or notification events currently exposed by the product |
Choose an integration¶
To invoke an existing operator from an application, publish an Operator API.
To let an agent discover and call tools from an MCP-compatible service, create an MCP tool.
To receive results or events without polling, configure a Webhook.
To run model inference rather than a data-processing operator, use the Genesis Model API.
For long-running, multi-step processing, orchestrate a workflow and use the job-query or callback capability available in your deployment. Do not treat a synchronous operator call as an asynchronous job.
Treat generated information as the contract¶
Extension behavior depends on the resource being published and on deployment configuration. This section therefore does not define a universal base URL, request schema, or signing algorithm. Build an integration against:
the endpoint, transport, and authentication method shown on the resource or configuration page;
the parameter reference, generated example, and test result for the selected version;
the release information and administrator settings for the current deployment.
After copying an example, replace its address and credentials, and never commit test credentials to source control. Recheck the contract and run an end-to-end test after changing an operator version, MCP service, or Webhook configuration.
For a broader capability map, see Platform Capabilities. For model authentication conventions, see Model API authentication.