Python: Add base_url parameter to AnthropicClient and RawAnthropicClient

[moonbox3]

Motivation and Context

Foundry users need to point AnthropicClient at a custom endpoint (e.g., Azure AI Foundry) just as the Anthropic SDK itself supports via base_url. Without this parameter, switching between Anthropic-hosted and Foundry-hosted deployments requires different client code, which is developer-unfriendly.

Fixes #5683

Description

AnthropicSettings TypedDict, RawAnthropicClient.__init__, and AnthropicClient.__init__ all lacked a base_url field. The fix adds base_url: str | None to AnthropicSettings and as a keyword argument to both client constructors, then threads it through load_settings and into the AsyncAnthropic(base_url=...) call, matching the pattern already used by AnthropicFoundryClient. The value also falls back to the ANTHROPIC_BASE_URL environment variable via the existing settings loader.

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.

---

Note: PR autogenerated by an agent

[moonbox3]

Automated Code Review

Reviewers: 4 | Confidence: 97% | Result: All clear

Reviewed: Correctness, Security Reliability, Test Coverage, Design Approach

---

Automated review by automated agents

[moonbox3]

Python Test Coverage Report •

File	Stmts	Miss	Cover	Missing
packages/anthropic/agent_framework_anthropic
_chat_client.py	442	35	92%	461, 464, 545, 638, 640, 783, 810–811, 889, 891, 921–922, 967, 983–984, 991–993, 997–999, 1003–1006, 1120, 1130, 1182, 1330–1331, 1348, 1361, 1374, 1399–1400
packages/core/agent_framework
_skills.py	592	10	98%	522, 532, 1808–1809, 1882, 1887, 1936, 1941, 2144–2145
TOTAL	33586	3908	88%

Python Unit Test Overview

Tests	Skipped	Failures	Errors	Time
6558	30 💤	0 ❌	0 🔥	1m 46s ⏱️

[github-actions]

Automated Code Review

Reviewers: 4 | Confidence: 90% | Result: All clear

Reviewed: Correctness, Security Reliability, Test Coverage, Design Approach

---

Automated review by moonbox3's agents

[Copilot]

Pull request overview

Adds configurable base_url support to the Python Anthropic clients so callers can target custom Anthropic-compatible endpoints (e.g., Azure AI Foundry) without changing client code, aligning behavior with the upstream Anthropic SDK and the existing Foundry-specific wrapper.

Changes:

• Add base_url: str | None to AnthropicSettings, and plumb it through RawAnthropicClient/AnthropicClient into AsyncAnthropic(base_url=...).
• Add unit tests verifying base_url passed as an explicit constructor kwarg is applied to the underlying SDK client.
• Minor formatting-only updates in skills-related core tests and error messages.

Reviewed changes

Copilot reviewed 5 out of 5 changed files in this pull request and generated 1 comment.

Show a summary per file

File	Description
python/packages/anthropic/agent_framework_anthropic/_chat_client.py	Adds base_url to settings + client constructors and passes it into AsyncAnthropic.
python/packages/anthropic/tests/test_anthropic_client.py	Adds tests for explicit base_url passthrough on both public and raw clients.
python/packages/core/agent_framework/_skills.py	Reflows a few exception messages (formatting-only).
python/packages/core/tests/core/test_skills.py	Formatting-only changes to test call sites/spacing.
python/packages/core/agent_framework/init.py	Minor __all__ ordering adjustment.

[moonbox3]

Automated Code Review

Reviewers: 4 | Confidence: 92%

✓ Correctness

The two new tests correctly exercise the ANTHROPIC_BASE_URL environment variable path for both AnthropicClient and RawAnthropicClient. The fixture machinery (override_env_param_dict → anthropic_unit_test_env) properly sets the env var via monkeypatch and returns it in the dict. The client's load_settings call uses env_prefix='ANTHROPIC_', so ANTHROPIC_BASE_URL maps to base_url in AnthropicSettings, which is then passed to AsyncAnthropic. No correctness issues found.

✓ Security Reliability

The diff adds two new unit tests verifying that AnthropicClient and RawAnthropicClient read ANTHROPIC_BASE_URL from the environment when no explicit base_url argument is provided. Both tests use the existing anthropic_unit_test_env fixture, which safely scopes env-var overides via monkeypatch. The conftest.py shows override_env_param_dict is merged into the env dict before monkeypatching, so the URL will be present. No injection risks, resource leaks, or reliability issues are present.

✓ Test Coverage

The two new tests correctly exercise the ANTHROPIC_BASE_URL env-var fallback path for both AnthropicClient and RawAnthropicClient. The fixture wiring is correct: override_env_param_dict with indirect=True is handled by conftest.py (returns request.param), anthropic_unit_test_env merges it into the base dict via env_vars.update(...) and calls monkeypatch.setenv for each key including ANTHROPIC_BASE_URL, so both the env var and the dict key are properly set. The assertion anthropic_unit_test_env["ANTHROPIC_BASE_URL"] is valid because the key is present in the returned dict after the update. Production code in _chat_client.py passes base_url=None into load_settings with env_prefix="ANTHROPIC_", which maps ANTHROPIC_BASE_URL → base_url correctly. One missing scenario: there is no test confirming that an explicitly-passed base_url takes precedence over a simultaneously set ANTHROPIC_BASE_URL env variable, which would close the priority-ordering contract.

✓ Design Approach

I did not find any design-approach issues in this diff. The added tests exercise the documented ANTHROPIC_BASE_URL fallback on both the raw and wrapped clients, and that behavior matches the existing load_settings(..., base_url=base_url) flow in agent_framework_anthropic/_chat_client.py where explicit arguments win and environment values fill in omitted settings.

Suggestions

• Consider adding a test that sets ANTHROPIC_BASE_URL in the environment and passes an explicit base_url kwarg, asserting the explicit value wins. This closes the priority-ordering contract (explicit arg > env var) that the existing tests leave implicit.

---

Automated review by automated agents

[moonbox3]

Automated Code Review

Reviewers: 4 | Confidence: 95% | Result: All clear

Reviewed: Correctness, Security Reliability, Test Coverage, Design Approach

---

Automated review by automated agents