fix: normalize Anthropic server tool usage

[Genmin]

Summary

• normalize dict server_tool_use values into ServerToolUse inside Usage
• preserve constructor/model_dump round-trips for raw Anthropic-compatible usage payloads
• cover the built-in tool cost tracking path that reads usage.server_tool_use.web_search_requests

Fixes #26749.

Tests

• /tmp/uv-0.10.9-env/bin/uv run --extra proxy pytest tests/test_litellm/types/test_types_utils.py tests/test_litellm/litellm_core_utils/llm_cost_calc/test_tool_call_cost_tracking.py -q
• /tmp/uv-0.10.9-env/bin/uv run --extra proxy ruff check --ignore PLR0915 litellm/types/utils.py tests/test_litellm/types/test_types_utils.py tests/test_litellm/litellm_core_utils/llm_cost_calc/test_tool_call_cost_tracking.py
• git diff --check

Note: full touched-file ruff still reports the existing PLR0915 limit on Delta.__init__; the targeted lint command above keeps the check focused on actionable findings for this patch.

[CLAassistant]

Thank you for your submission! We really appreciate it. Like many open source projects, we ask that you sign our Contributor License Agreement before we can accept your contribution.
_{You have signed the CLA already but the status is still pending? Let us recheck it.}

[greptile-apps]

Greptile Summary

This PR fixes a bug where Anthropic passthrough responses that construct Usage from a raw dict payload would fail downstream when StandardBuiltInToolCostTracking tried to read usage.server_tool_use.web_search_requests — because the dict was never coerced into a ServerToolUse instance. The fix normalizes dict values to ServerToolUse inside Usage.__init__, mirrors the existing pattern already used for prompt_tokens_details and completion_tokens_details, and is covered by two new unit tests including a model_dump() round-trip check.

Confidence Score: 5/5

Safe to merge — targeted, well-tested fix with no backwards-incompatible changes.

The change is minimal and follows an established pattern already used for other nested usage fields. Both the conversion logic and the round-trip are directly tested. Unused imports (including a rule-violating fastapi.testclient import) are cleaned up as a side effect.

No files require special attention.

Important Files Changed

Filename	Overview
litellm/types/utils.py	Adds dict-to-ServerToolUse normalization in Usage.init; field annotation stays typed as Optional[ServerToolUse] and conversion happens correctly after super().init().
tests/test_litellm/litellm_core_utils/llm_cost_calc/test_tool_call_cost_tracking.py	Adds test covering the dict server_tool_use path through StandardBuiltInToolCostTracking; removes unused imports including the fastapi TestClient import that violated the no-FastAPI-outside-proxy rule.
tests/test_litellm/types/test_types_utils.py	Adds test for dict-to-ServerToolUse conversion and model_dump round-trip; cleans up unused imports.

_{Reviews (1): Last reviewed commit: "fix: normalize Anthropic server tool usa..." | Re-trigger Greptile}

[codspeed-hq]

Merging this PR will not alter performance

✅ 16 untouched benchmarks

---

_{Comparing Genmin:fix/anthropic-passthrough-server-tool-use (380214f) with main (3e1479c)}

[codecov]

Codecov Report

✅ All modified and coverable lines are covered by tests.

📢 Thoughts on this report? Let us know!