About the Responses API and Agent#

The NVIDIA NeMo Agent toolkit supports OpenAI’s Responses API through two complementary pieces:

  • Configuring the LLM client mode using the api_type field

  • Integrating tool binding with the NeMo Agent toolkit dual-node graph using the dedicated workflow agent _type: responses_api_agent, designed for tool use with the Responses API.

The Responses API enables models to:

  • Use built-in tools such as Code Interpreter through builtin_tools.

  • Connect to remote tools using Model Context Protocol (MCP) through mcp_tools, specifying fields such as server_label and server_url.

  • Use toolkit tools through nat_tools, executed by the agent graph.

To configure your LLM agent for the Responses API and use the dedicated agent, refer to Configure the Responses API and Agent.