Start the Microsoft Agent Framework Bot on an Anthropic key

[kevin9327]

What this changes

Picked with an Anthropic key, the Microsoft Agent Framework Bot exits on startup.

The harness built OpenAIChatClient(BOT_MODEL) whatever the setup screen chose, and never read BOT_PROVIDER. The desktop catalogue gives every harness it installs Credential::AnyProvider, and this row's own summary is "Microsoft's, model-agnostic by design", so it can be paired with an Anthropic key. The desktop then writes BOT_PROVIDER=anthropic and BOT_MODEL=claude-sonnet-4-5, and leaves OPENAI_API_KEY empty. The OpenAI client refuses to be built without a key, and it is built at module level, so the import failed and the container never served /health:

openai.OpenAIError: Missing credentials. Please pass an `api_key`, ... or set the `OPENAI_API_KEY` ...

Agent Framework has its own Anthropic client. AnthropicClient is published as agent-framework-anthropic and re-exported from agent_framework.anthropic, the same way agent_framework.openai re-exports the OpenAI one. It reads ANTHROPIC_API_KEY and ANTHROPIC_BASE_URL. The harness now builds that client when the provider is anthropic, and OpenAIChatClient as before otherwise, so the OpenAI key and OpenAI-compatible endpoint choices are unchanged.

One thing to weigh: agent-framework-anthropic has only ever been published as a pre-release. Its newest is 1.0.0b260918, and PyPI lists no final version. pip installs it from a plain requirement because nothing else satisfies it, which is what the Linux run below did. agent-framework-openai and agent-framework-ag-ui are final releases. Google is left out: the setup screen offers no Google key, and Compose passes the harness no GOOGLE_API_KEY.

Where it runs

• New state that outlives a request? None. The client is chosen once at import.
• What happens on the second replica? The same. Each harness process picks the same client from the same environment.
• Anything serialised? No.
• Anything fanned out to a browser? No.
• New listener, port, or schedule? No. The CI step added here runs beside the harness suites already in python-harness.

Boundary and audit

• Every acting call still goes through the gateway: nothing on the server changes, and the harness still refuses a request without the server's token.
• New refusals and new failures each write a row: no new refusal or failure.
• Nothing new is trusted from the client.

Changelog

• A line in CHANGELOG.md under Unreleased.

Proof

• New agent-microsoft/tests/test_main.py follows agent-llamaindex/tests/test_main.py. It starts a local fake provider, imports the harness under the environment the desktop writes for each of the three choices, and posts an AG-UI run to /. It asserts HTTP 200, RUN_FINISHED, no RUN_ERROR, the provider's answer in the stream, and that the provider got the chosen model on the chosen route. The fake serves /v1/responses because OpenAIChatClient is Agent Framework's Responses API client. It also serves /v1/messages.
• Linux (ubuntu-latest, Python 3.12, a throwaway workflow on my fork running the same commands as the new CI step). This branch: 3 passed. With agent-microsoft/src from main (2d09a08): 1 failed, 2 passed. [an Anthropic key] fails at import with the OpenAIError above.
• Windows, Python 3.12: the same results.
• Versions pip resolved: agent-framework-core 1.19.0, agent-framework-openai 1.14.4, agent-framework-anthropic 1.0.0b260918, agent-framework-ag-ui 1.4.0, anthropic 0.116.0.
• Not run: a real Anthropic key, and the published image.

This adds a step at the end of python-harness and an entry at the top of Unreleased, which #609 to #612 also touch. Whichever lands later needs a rebase, and I'll do it.

Default endpoint regression

The default Compose environment sets ANTHROPIC_BASE_URL to an empty string. Microsoft Agent Framework's Anthropic client passed that through instead of selecting the official endpoint. The harness now explicitly uses https://api.anthropic.com for a blank override and preserves a nonblank custom URL.

The additional authenticated AG-UI regression uses the real framework and Anthropic SDK, replacing only HTTP transport. It checks the official URL, API-key header, chosen model, answer, and successful run completion. All four tests pass locally, including the original provider cases. This is SDK integration coverage; no live Anthropic account or published image was exercised.

🤖 Generated with Claude Code

[chatgpt-codex-connector]

You have reached your Codex usage limits for code reviews. You can see your limits in the Codex usage dashboard.