MCP Call Arguments has wrong chunk type

API Bugs
,
1

Hi everyone,

I’ve been experimenting with MCP in the Responses API, and I believe that for some messages, the type chunk is incorrect.

With this code:

❯ curl https://api.openai.com/v1/responses \
  -H "Content-Type: application/json" \
  -H "Authorization: Bearer $OPENAI_API_KEY" \
  -d '{
  "model": "o4-mini",
    "reasoning": {"effort": "medium", "summary": "detailed"},
  "tools": [
    {
      "type": "mcp",
      "server_label": "deepwiki",
      "server_url": "https://mcp.deepwiki.com/mcp",
      "require_approval": "never"
    }
  ],
  "input": "What transport protocols are supported in the 2025-03-26 version of the MCP spec? Before you call any tool, tell me if youre calling a tool any why. FYI: the github is modelcontextprotocol/modelcontextprotocol",
  "stream": true
}'

I get some events looking like this:

    {
        "type": "response.mcp_call_arguments.delta",
        "sequence_number": 10,
        "output_index": 2,
        "item_id": "mcp_684ff3a22a60819c90fa205XXXXXXXXXXX",
        "delta": "{\"repoName\":\"modelcontextprotocol/modelcontextprotocol\"}"
    },
    {
        "type": "response.mcp_call_arguments.done",
        "sequence_number": 11,
        "output_index": 2,
        "item_id": "mcp_684ff3a22a60819c90fa205XXXXXXXXXXX",
        "arguments": "{\"repoName\":\"modelcontextprotocol/modelcontextprotocol\"}"
    },

The two types, here, are response.mcp_call_arguments.delta and response.mcp_call_arguments.done. But in the API reference, they should be response.mcp_call.arguments.delta and response.mcp_call.arguments.done, respectively. (See https://platform.openai.com/docs/api-reference/responses-streaming/response/mcp_call/arguments/delta)

This is not about being pedantic: this breaks the Python SDK, as far as I can tell:

from openai.types import responses as rt # version 1.86.0
from pydantic import TypeAdapter

attempt = [
    {
        "type": "response.mcp_call_arguments.delta",
        "sequence_number": 10,
        "output_index": 2,
        "item_id": "mcp_684ff3a22a60819c90fa205XXXXXXXXXXX",
        "delta": '{"repoName":"modelcontextprotocol/modelcontextprotocol"}',
    },
    {
        "type": "response.mcp_call_arguments.done",
        "sequence_number": 11,
        "output_index": 2,
        "item_id": "mcp_684ff3a22a60819c90fa205XXXXXXXXXXX",
        "arguments": '{"repoName":"modelcontextprotocol/modelcontextprotocol"}',
    },
]
_ = TypeAdapter(list[rt.ResponseStreamEvent]).validate_python(attempt) # Will not work

Now, funnily enough, even with the correct type, the code will not work, because the Python SDK also has the wrong type. (PR on the Python SDK GitHub repo is 2412.) But the point about the API being deficient stands.

Thanks :slight_smile: