The `/tools/list` endpoint lets an MCP server declare its available tools. It returns machine-readable descriptions including each tool’s name, input parameters, output format, and purpose. section 7.6.1
When an MCP server is added to the tools array, the OpenAI runtime queries the endpoint at the start of the conversation and loads the schemas into context. The model can then reason about the tools like native functions, without manually supplying schemas inline. section 7.6.1
Describing available actions
A server might expose semantically named actions such as “search for a product” and “check availability.” section 7.6.1
Figure 7.2 How AI discovers and uses MCP tools. When tools are registered on the server, the AI model can automatically discover, reason about, and invoke them.
Shows the flow from registering server tools to model discovery and invocation. section 7.6.1
