Python: feat(foundry): add experimental hosted tool factories on FoundryChatClient#5958
Conversation
…lient Adds eight new `@experimental` static factory methods on `FoundryChatClient` covering Foundry-hosted tools that previously had no helper: - get_azure_ai_search_tool - get_sharepoint_tool - get_fabric_tool - get_memory_search_tool - get_computer_use_tool - get_browser_automation_tool - get_bing_custom_search_tool - get_a2a_tool All factories are marked with the new `ExperimentalFeature.FOUNDRY_TOOLS` tag and resolve the underlying `azure-ai-projects` preview classes lazily through a `_require_sdk_class` helper so older SDK versions still import cleanly and fail with a clear `ImportError` only on use. Tests cover each factory's return type and field wiring, the experimental metadata, and the missing-SDK-class fallback. Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>
There was a problem hiding this comment.
Pull request overview
Note
Copilot was unable to run its full agentic suite in this review.
Adds experimental, instance-free hosted tool factory helpers to FoundryChatClient, with lazy resolution of preview SDK classes to preserve import compatibility with older azure-ai-projects versions.
Changes:
- Added
_require_sdk_class()helper to lazily resolve preview tool classes and raise a clearImportErrorwhen missing. - Added eight new
@experimentalstatic hosted-tool factories (Azure AI Search, SharePoint, Fabric, Memory Search, Computer Use, Browser Automation, Bing Custom Search, A2A). - Added unit tests for the new factories and documented the new factory list + experimental feature flag in the README.
Reviewed changes
Copilot reviewed 4 out of 4 changed files in this pull request and generated 4 comments.
| File | Description |
|---|---|
| python/packages/foundry/agent_framework_foundry/_chat_client.py | Adds lazy SDK-class resolver and eight experimental hosted tool factories. |
| python/packages/foundry/tests/foundry/test_foundry_chat_client.py | Adds unit tests validating factory output shapes, experimental metadata, and missing-class error behavior. |
| python/packages/foundry/README.md | Documents GA vs experimental hosted-tool factories and the lazy SDK resolution behavior. |
| python/packages/core/agent_framework/_feature_stage.py | Introduces ExperimentalFeature.FOUNDRY_TOOLS feature id. |
* Skip preview-tool tests gracefully (`_skip_if_sdk_class_missing`) when
the installed `azure-ai-projects` does not expose the required preview
class, matching the lazy-import guard in production code so the test
suite stays green on older SDK installs.
* Add `filterwarnings("ignore::FutureWarning")` to each new tool-factory
test (and the parametrized metadata test) so they remain stable under
strict warning configurations \u2014 the global dedup in
`_feature_stage._WARNED_FEATURES` makes `pytest.warns` brittle across
ordered runs.
* Use `monkeypatch.setattr(..., None, raising=False)` instead of
`delattr` in the missing-SDK-class test so it works for modules that
implement PEP 562 `__getattr__`.
* Split the long `get_bing_custom_search_tool` return into two lines for
readability.
Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>
github-actions
Bot
changed the title
|
All four Copilot review comments are addressed in The CI failures on this PR are unrelated to the changes — every failing job (pre-commit, mypy, samples, package checks, tests on 3.13/windows) dies in the shared
The |
![github-actions[bot]](https://avatars.githubusercontent.com/in/15368?s=60&v=4){kind=small}
There was a problem hiding this comment.
Automated Code Review
Reviewers: 4 | Confidence: 86%
✓ Correctness
The PR is well-implemented with no correctness bugs found. The factory methods correctly construct nested SDK objects, the lazy import helper properly surfaces ImportError for missing SDK symbols, decorator interaction with
@staticmethodpreserves metadata as expected, and tests are thorough with proper warning suppression. The existing review thread items have been addressed (monkeypatch uses setattr with None, tests have filterwarnings marks).
✓ Security Reliability
The PR adds eight experimental Foundry tool factories with a lazy SDK class resolution helper. The implementation is generally sound, but three factories (
get_azure_ai_search_tool,get_memory_search_tool,get_bing_custom_search_tool) construct parameter dicts where**kwargsis spread after explicit keys, allowing kwargs to silently override security-sensitive values likeproject_connection_id. This is inconsistent withget_a2a_toolin the same PR which correctly puts kwargs as the base and applies explicit params on top.
✓ Test Coverage
Test coverage for the new experimental hosted tool factories is thorough. Each of the 8 new factories has a dedicated happy-path test with meaningful assertions on nested structure, a parametrized test verifies experimental metadata, an error-path test covers SDK class resolution failure, and all tests gracefully skip on older SDK versions. The filterwarnings marks correctly handle ExperimentalWarning (subclass of FutureWarning). No significant coverage gaps found.
✓ Design Approach
I did not find a blocking design-approach problem in the factory implementations themselves. The one issue worth tightening is the README’s warning semantics: the new factories all share the same
ExperimentalFeature.FOUNDRY_TOLS, and the feature-stage system emits warnings once per feature ID, so the docs currently over-promise a per-factory warning behavior.
Automated review by eavanvalkenburg's agents
* Reorder the dict-literal kwargs assembly in get_azure_ai_search_tool, get_memory_search_tool, and get_bing_custom_search_tool so explicit parameters always take precedence over **kwargs (matching the safe pattern already used in get_a2a_tool). This prevents a caller passing `project_connection_id`, `index_name`, `memory_store_name`, `scope`, or `instance_name` through `**kwargs` from silently overriding the explicit security-sensitive arguments. * Update the README experimental note to reflect once-per-feature-id dedup semantics of `_feature_stage._WARNED_FEATURES` rather than claiming a per-factory "first use" warning. Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>
Motivation and Context
Agent Framework's
FoundryChatClientexposes static helpers (get_code_interpreter_tool,get_file_search_tool,get_web_search_tool,get_image_generation_tool,get_mcp_tool) for building Foundry-hosted tool configurations without needing a client instance. The Foundry SDK ships several more hosted tools — Azure AI Search, SharePoint, Fabric, Memory Search, Computer Use, Browser Automation, Bing Custom Search, A2A — and there was no equivalent helper for any of them, so developers had to reach intoazure.ai.projects.modelsdirectly and assemble the nested parameter types by hand.This PR fills that gap and lays groundwork for an upcoming
to_prompt_agent()converter (separate PR) that wants a portable, instance-free way to construct any Foundry-hosted tool.Description
@experimentalstatic factories onFoundryChatClient, one per missing hosted tool. Each wraps the matchingazure.ai.projects.modelstype (AzureAISearchTool,SharepointPreviewTool,MicrosoftFabricPreviewTool,MemorySearchPreviewTool,ComputerUsePreviewTool,BrowserAutomationPreviewTool,BingCustomSearchPreviewTool,A2APreviewTool) and hides the nested parameter/resource shapes behind a flat, keyword-only signature.ExperimentalFeature.FOUNDRY_TOOLSfeature id (the existing GA factories are unchanged)._require_sdk_class()helper that resolves the preview classes fromazure.ai.projects.modelslazily and raises a clearImportErrorif the installedazure-ai-projectsis too old to expose them — older SDK installs continue to import the package fine and only fail when an unsupported factory is actually called.ExperimentalFeature.FOUNDRY_TOOLStag.Contribution Checklist