[Somay-kousis]

Link to Issue or Description of Change

1. Link to an existing issue (if applicable):

• Closes: AnthropicLlm ignores tool_config: function calling mode NONE or ANY never reaches Claude #7633

2. Or, if no issue exists, describe the change:

Problem:

AnthropicLlm always sent tool_choice auto, so the function calling mode in generate_content_config.tool_config never reached Claude. NONE did not stop tool calls, ANY did not force one, and the model_consult force plugin (ANY with one allowed name) had no effect on Claude.

Solution:

AnthropicLlm now maps config.tool_config.function_calling_config to Claude's tool_choice:

ADK mode	Claude tool_choice
NONE	{"type": "none"}
ANY with one allowed_function_names entry	{"type": "tool", "name": ...}
ANY otherwise	{"type": "any"}
AUTO, VALIDATED, unset, no config	{"type": "auto"} (unchanged)

Two limits come from the Anthropic API:

• Claude rejects forced tool use (any or tool) while extended thinking is on. In that case ANY falls back to auto with a warning, the same way the existing output-schema tool falls back. NONE is kept, since Claude accepts it with thinking.
• Claude can't limit the call to a subset of several tools, so two or more allowed names map to any. integrations/openai/_openai_common.py handles the same case this way.

Requests with no tools still send no tool_choice. The output-schema path, which sets its own choice, is unchanged.

Testing Plan

Unit Tests:

• I have added or updated unit tests for my change.
• All unit tests pass locally.

New tests in tests/unittests/models/test_anthropic_llm.py:

• test_tool_config_mode_reaches_claude_tool_choice: every mode, streaming and non-streaming.
• test_forced_tool_choice_with_thinking_falls_back_to_auto: ANY and ANY with one name, with a manual budget and with adaptive thinking.
• test_tool_config_none_mode_is_kept_with_thinking.

main: 9 failed, 193 passed
fix:  202 passed

The 10 new cases that pass on main are the ones where auto is correct. Three mutations of the fix (NONE back to auto, single-name branch disabled, thinking fallback removed) each make at least 3 tests fail. tests/unittests/models and tests/unittests/flows pass (2814 tests). isort, pyink and mypy are clean on the changed file.

Manual End-to-End (E2E) Tests:

The repro in the issue (Agent + InMemoryRunner, AsyncMessages.create faked to record the request):

before                                   after
NONE None -> {'type': 'auto'}            NONE None -> {'type': 'none'}
ANY None -> {'type': 'auto'}             ANY None -> {'type': 'any'}
ANY ['get_weather'] -> {'type': 'auto'}  ANY ['get_weather'] -> {'type': 'tool', 'name': 'get_weather'}

No live Claude call; the thinking limits come from Anthropic's tool use docs and the fallback this file already has.

Checklist

• I have read the CONTRIBUTING.md document.
• I have performed a self-review of my own code.
• I have commented my code, particularly in hard-to-understand areas.
• I have added tests that prove my fix is effective or that my feature works.
• New and existing unit tests pass locally with my changes.
• I have manually tested my changes end-to-end.
• Any dependent changes have been merged and published in downstream modules.

Additional context

LiteLlm also ignores a single allowed function name (ANY always becomes "required", lite_llm.py:3645-3654). Not changed here.