Python: drop hosted MCP calls when reasoning is stripped

[he-yufeng]

Motivation and Context

Fixes #6074.

When OpenAI Responses history is replayed without service-side storage, the client strips text_reasoning because replaying standalone reasoning items is rejected by the API. Hosted MCP calls from the same model output were still kept, which can leave the request with a bare mcp_call after its paired reasoning item has been removed.

Description

This change treats that stateless replay path as an all-or-nothing pair for hosted MCP items. If a message contains reasoning that will be stripped, hosted MCP call/result contents in the same message are skipped too. Any later unmatched MCP result marker is then dropped by the existing coalescing logic.

The existing behavior for hosted MCP calls without reasoning is unchanged, and storage-backed continuation still strips server-issued MCP items as before.

A regression test covers a reasoning plus hosted MCP call/result history and verifies that the prepared input does not contain a bare reasoning, mcp_call, or function_call_output item.

Contribution Checklist

• The code builds clean without any errors or warnings
• The PR follows the Contribution Guidelines
• All unit tests pass, and I have added new tests where possible
• Is this a breaking change? If yes, add "[BREAKING]" prefix to the title of the PR.

[Copilot]

Pull request overview

Note

Copilot was unable to run its full agentic suite in this review.

Adds handling for a replay scenario where reasoning content is stripped (when not using service-side storage) by also dropping the paired hosted-MCP tool call/result to avoid emitting invalid item sequences to the OpenAI API.

Changes:

• Add a unit test that asserts hosted-MCP call/result are dropped when paired reasoning is stripped (no service-side storage).
• Update _prepare_message_for_openai to detect “reasoning will be dropped” and skip serializing mcp_call accordingly.

Reviewed changes

Copilot reviewed 2 out of 2 changed files in this pull request and generated 2 comments.

File	Description
python/packages/openai/tests/openai/test_openai_chat_client.py	Adds regression test for dropping hosted-MCP call/result when reasoning is stripped.
python/packages/openai/agent_framework_openai/_chat_client.py	Drops mcp_call when reasoning is stripped without storage to prevent invalid payloads.

[moonbox3]

Python Test Coverage Report •

File	Stmts	Miss	Cover	Missing
packages/openai/agent_framework_openai
_chat_client.py	1092	148	86%	276, 289, 631–635, 643–646, 652–656, 706–713, 715–717, 724–726, 772, 780, 803, 921, 1020, 1079, 1081, 1083, 1085, 1151, 1165, 1245, 1255, 1260, 1303, 1414–1415, 1430, 1657, 1662, 1666–1668, 1672–1673, 1756, 1766, 1793, 1799, 1809, 1815, 1820, 1826, 1831–1832, 1851, 1854–1857, 1871, 1873, 1881–1882, 1894, 1936, 2026, 2048–2049, 2064–2065, 2083–2084, 2127, 2293, 2331–2332, 2350, 2430–2438, 2468, 2578, 2613, 2628, 2648–2658, 2671, 2682–2686, 2700, 2714–2725, 2734, 2766–2769, 2779–2780, 2791–2793, 2807–2809, 2819–2820, 2826, 2841
TOTAL	37988	4426	88%

Python Unit Test Overview

Tests	Skipped	Failures	Errors	Time
7579	34 💤	0 ❌	0 🔥	1m 53s ⏱️

[moonbox3]

Please check the failing tests as well.

[he-yufeng]

Rebased this branch onto latest upstream/main and re-ran the checks for the OpenAI changes here.

Local validation:

• uv run --dev python -m ruff check packages/openai/agent_framework_openai/_chat_client.py packages/openai/tests/openai/test_openai_chat_client.py
• uv run --dev python -m py_compile packages/openai/agent_framework_openai/_chat_client.py packages/openai/tests/openai/test_openai_chat_client.py
• uv run --dev python -m mypy --config-file pyproject.toml packages/openai/agent_framework_openai/_chat_client.py
• uv run --dev python -m pytest packages\openai\tests\openai\test_openai_chat_client.py -q -m "not integration" --basetemp .tmp\pytest-6210-unit

I also checked the failed GitHub Actions job. The failing cases are the Docker shell-tool tests, and the failure is from pulling alpine:3:

Unable to find image 'alpine:3' locally
docker: Error response from daemon: Get "https://registry-1.docker.io/v2/": context deadline exceeded

Those tests do not touch the OpenAI message serialization path changed in this PR. I did not change Docker-related code.

One local note: running the full test_openai_chat_client.py file includes live integration tests; those are blocked in my local environment by missing model config / an invalid local OpenAI credential, so I used the non-integration subset above for this PR's regression coverage.

[he-yufeng]

Follow-up pushed in 207e60c89.

Changes:

• lifted the hosted-MCP drop decision from per-message to the full replay input, so custom history providers that split text_reasoning, mcp_server_tool_call, and mcp_server_tool_result across adjacent messages do not replay a bare mcp_call;
• added a regression test for that cross-message case;
• kept ordinary function_call replay unchanged under store=False.

I checked the function_call question, and it is not equivalent to hosted MCP in this path: the existing store=false tool-loop tests need the function_call and matching function_call_output to be replayed together on the second request. Dropping them breaks test_tool_loop_store_false_omits_reasoning_items_from_second_request and the full-conversation reasoning/function-call test. Hosted MCP is different here because the call/result is serialized as one mcp_call item, and replaying the call side without its stripped reasoning is the invalid shape this PR is fixing.

Validation:

uv run --dev python -m pytest packages\openai\tests\openai\test_openai_chat_client.py -q -m "not integration" --basetemp .tmp\pytest-6210-20260604b
uv run --dev python -m ruff check packages\openai\agent_framework_openai\_chat_client.py packages\openai\tests\openai\test_openai_chat_client.py
uv run --dev python -m py_compile packages\openai\agent_framework_openai\_chat_client.py packages\openai\tests\openai\test_openai_chat_client.py
uv run --dev python -m mypy --config-file pyproject.toml packages\openai\agent_framework_openai\_chat_client.py
git diff --check upstream/main..HEAD