Python: Add Mistral chat client

[orangeCatDeveloper]

Motivation & Context

Issue #7366 asks for a native Mistral chat client so agents can run on Mistral models with first-class support, instead of having no Python path at all (the package previously only shipped an embedding client). This PR implements it for the Python SDK.

Description & Review Guide

• What are the major changes?
  • New MistralChatClient / RawMistralChatClient in agent-framework-mistral with streaming, tool calling, structured output, multimodal input, and reasoning content, plus a runnable agent sample and docs.
  • Both the new chat client and the existing embedding client talk to the Mistral REST API directly over httpx instead of the mistralai SDK: the SDK pins opentelemetry-semantic-conventions<0.61, which is unsatisfiable next to the workspace's OpenTelemetry stack (pinned to 0.64b0 by microsoft-opentelemetry). Same approach as LangChain's langchain-mistralai and LiteLLM; keeps the package in core[all] with a clean pip check.
  • Integration tests wired into the misc CI job, skipping gracefully when MISTRAL_API_KEY is unset.
• What is the impact of these changes?
  • from agent_framework.mistral import MistralChatClient now works; agent-framework[all] includes Mistral again. No changes to other packages' behavior; the mistralai dependency is dropped in favor of httpx.
• What do you want reviewers to focus on?
  • Streaming tool-call handling: Mistral may split IDs/names/arguments across chunks and interleave parallel calls, and omits IDs on some paths (null/"null"), so fragments are accumulated per (choice, index) and emitted on the finish chunk. Also La Plateforme's 9-char alphanumeric tool-call ID requirement is handled via deterministic sanitization.

Verified: 72 unit tests over httpx.MockTransport asserting real wire JSON; mypy/pyright strict clean; live runs against api.mistral.ai covering chat, streaming, agent tool loop, structured output, embeddings (incl. output_dimension on codestral-embed), and both samples.

Note for maintainers: running the CI integration tests requires the MISTRAL_API_KEY secret and MISTRAL_CHAT_MODEL_ID/MISTRAL_EMBEDDING_MODEL_ID repo vars; without them the tests skip.

Related Issue

Fixes #7366

Contribution Checklist

• The code builds clean without any errors or warnings
• All unit tests pass, and I have added new tests where possible
• The PR follows the Contribution Guidelines
• This PR is linked to an issue and there is no other open PR for this issue (see Related Issue above).
• This is not a breaking change. If it is a breaking change, add the breaking change label (or add "[BREAKING]" to the title prefix, before or after any language prefix) — a workflow keeps the label and title prefix in sync automatically.

[orangeCatDeveloper]

@orangeCatDeveloper please read the following Contributor License Agreement(CLA). If you agree with the CLA, please reply with the following information.

@microsoft-github-policy-service agree [company="{your company}"]

Options:

• (default - no company specified) I have sole ownership of intellectual property rights to my Submissions and I am not making Submissions in the course of work for my employer.

@microsoft-github-policy-service agree

• (when company given) I am making Submissions in the course of work for my employer (or my employer has intellectual property rights in my Submissions by contract or applicable law). I have permission from my employer to make Submissions and enter into this Agreement on behalf of my employer. By signing below, the defined term “You” includes me and my employer.

@microsoft-github-policy-service agree company="Microsoft"

Contributor License Agreement

@microsoft-github-policy-service agree

[Copilot]

Pull request overview

Adds a native Python MistralChatClient to the agent-framework-mistral integration, and switches both chat + embeddings to call the Mistral REST API directly via httpx (removing the mistralai SDK dependency to avoid OpenTelemetry version conflicts). This expands the Mistral provider feature set (streaming, tool calling, structured output, multimodal inputs, reasoning content), and wires the package into CI test runs.

Changes:

• Implement MistralChatClient / RawMistralChatClient using httpx + SSE streaming, including tool-call ID sanitization and streamed tool-call fragment coalescing.
• Rework MistralEmbeddingClient to use direct REST calls over httpx and update unit/integration tests accordingly.
• Add/refresh runnable samples + docs, and include Mistral tests + secrets/vars wiring in Python CI workflows.

Reviewed changes

Copilot reviewed 19 out of 20 changed files in this pull request and generated 3 comments.

Show a summary per file

File	Description
python/uv.lock	Drops mistralai and locks httpx as the Mistral package dependency.
python/samples/README.md	Documents required Mistral environment variables for samples.
python/samples/02-agents/providers/mistral/README.md	Updates provider sample README to include chat + agent sample.
python/samples/02-agents/providers/mistral/mistral_embeddings.py	Updates embeddings sample to close the client and clarifies options.
python/samples/02-agents/providers/mistral/mistral_agent_basic.py	Adds a runnable agent sample demonstrating tool use + streaming.
python/packages/mistral/tests/mistral/test_mistral_embedding_client.py	Replaces SDK-mocking tests with httpx.MockTransport wire-JSON tests and adds error cases.
python/packages/mistral/tests/mistral/test_mistral_chat_client.py	Adds comprehensive unit + integration coverage for the new chat client (streaming, tools, structured output, errors).
python/packages/mistral/README.md	Adds chat client documentation and refreshes embedding docs with proper client closing.
python/packages/mistral/pyproject.toml	Replaces mistralai dependency with httpx.
python/packages/mistral/AGENTS.md	Updates package documentation to include chat client and REST-over-httpx implementation notes.
python/packages/mistral/agent_framework_mistral/_embedding_client.py	Reimplements embeddings via REST + httpx, adds close semantics and improved error wrapping.
python/packages/mistral/agent_framework_mistral/_chat_client.py	New REST-based chat client with streaming SSE parsing, tool calling, structured output, and telemetry/middleware layers.
python/packages/mistral/agent_framework_mistral/init.py	Exports new chat client types and settings.
python/packages/core/tests/core/test_mistral_namespace.py	Ensures the agent_framework.mistral namespace exposes the new Mistral chat exports.
python/packages/core/agent_framework/mistral/init.pyi	Updates type-stub exports for the new Mistral chat types.
python/packages/core/agent_framework/mistral/init.py	Adds lazy-import mapping for the new Mistral chat exports.
python/.env.example	Adds MISTRAL_CHAT_MODEL and MISTRAL_SERVER_URL examples.
.github/workflows/python-merge-tests.yml	Includes Mistral package tests in the “misc” Python workflow and wires Mistral secrets/vars.
.github/workflows/python-integration-tests.yml	Includes Mistral package tests in integration workflow and wires Mistral secrets/vars.
.github/workflows/integration-tests-manual.yml	Adds MISTRAL_API_KEY secret to the manual integration workflow environment.

[eavanvalkenburg]

Please make sure to use the defined PR template @orangeCatDeveloper

[github-actions]

Python Test Coverage Report •

File	Stmts	Miss	Cover	Missing
packages/mistral/agent_framework_mistral
_chat_client.py	458	0	100%
_embedding_client.py	93	0	100%
TOTAL	46074	4477	90%

Python Unit Test Overview

Tests	Skipped	Failures	Errors	Time
9471	34 💤	0 ❌	0 🔥	2m 23s ⏱️

[eavanvalkenburg]

I'm not sure I agree wholeheartedly with not using the mistral package, it tends to ensure a bit better behavior, but I will look into that. Key thing to focus on is all the handling of function calls, most of that should already be covered by the framework, so there is no need to do that inside a provider in this way.

[orangeCatDeveloper]

Thanks for the review! All feedback addressed in latest commit