An inference and tool interface
The Responses API generates model outputs and supports tool use. With direct Responses integration, your application owns the orchestration loop. Tool-capable inference is a building block, not the same resource as a managed Agents API session.
Own the surrounding workflow
Decide how to retain conversation state, execute custom functions, enforce permissions, and recover from failure. Hosted tools can reduce integration work, but your application still decides how the overall workflow proceeds.
Avoid schema confusion
Managed Agents API MCP tools use their documented transport configuration. Do not paste a Responses tool object into an Agents API session without checking its schema. Event names, state resources, and cleanup behavior differ too.
Read the official reference
Check the source for current API fields, account requirements, and service limits.
OpenAI: agents