WDTS — native Copilot MCP eligibility, current certification schema and Azure boundary

Question

WDTS — native Copilot MCP eligibility, current certification schema and Azure boundary

jeff ostler 0

Hello Microsoft team,

We are preparing What Do They Say?, an OAuth-protected MCP service for approved comparisons of independent AI answers. We want the user's actual preferred Copilot conversation to remain Main AI. The person selects a brief and an original answer; after exact approval, WDTS sends only that brief to selected other labs' official APIs. The Main AI original and other responses remain separate until the complete packet returns to Copilot for analysis. Saving the unchanged analysis requires separate permission.

Our current hosted integration targets ChatGPT and Claude; Microsoft compatibility is not implemented or tested. Please clarify:

Which consumer Copilot or Microsoft 365 Copilot hosts support this in an existing conversation? Does an agent connector preserve that context, or would a separate agent be required? We do not want to imply an Azure-hosted replacement chat meets the same experience.
Which current Partner Center route and artifact should we use: MCP agent connector, declarative-agent plugin, or another package? What account, publisher, tenant, license, region and mobile requirements apply to an Australian sole trader?
The MCP certification page requires AzureKeyVault and says DCR is unsupported, while the OAuth guide permits DCR and says AzureKeyVault fails numbered-schema validation. Which exact schema/auth configuration applies to our intended public route? Please also confirm required intro.md, callback/client registration and credential-delivery arrangements. We will not put secrets in the package.
Can a short job acknowledgement followed by status and bounded result-part retrieval meet review timing and input/output limits? We must detect truncation and return all originals before claiming complete analysis.
What rules cover selected original-answer transfer, saving analysis, external lab processing and interoperability wording? Which prepaid/subscription-included/sponsored allowance profiles are permitted, and what fees or revenue-sharing arrangements apply?

Please distinguish native Copilot distribution from Azure Foundry hosting and Azure Marketplace commerce. We understand Azure is a platform, not evidence of another independent model lab. We do not need to move hosting or buy services merely to establish the right integration route.

We can provide the data flow, source-derived tool catalogue and synthetic review cases. This is a presubmission route and eligibility enquiry, not a completed certification package or a request to publish.

Thank you,

Jeff

What Do They Say?

https://whatdotheysay.app

jeff ostler 0 Reputation points

2026-10-12T03:56:43.3866667+00:00

Following up with a narrower starting point: our first proposed Microsoft test would retrieve only authored saved-answer samples and optionally save the user-approved analysis, with no new model calls or paid-allowance deductions. Could the Copilot extensibility team confirm which current consumer or Microsoft 365 host can support this within the person's existing conversation, and which exact package/authentication schema should be used? We have private ChatGPT and Claude sample tests, not Microsoft compatibility or certification. If this needs a different official review channel, please direct us there. Publisher identity/business verification is separate from this technical eligibility question. Thank you, Jeff.

1 answer

Your answer

jeff ostler 0 Reputation points

2026-10-12T03:56:43.3866667+00:00

Following up with a narrower starting point: our first proposed Microsoft test would retrieve only authored saved-answer samples and optionally save the user-approved analysis, with no new model calls or paid-allowance deductions. Could the Copilot extensibility team confirm which current consumer or Microsoft 365 host can support this within the person's existing conversation, and which exact package/authentication schema should be used? We have private ChatGPT and Claude sample tests, not Microsoft compatibility or certification. If this needs a different official review channel, please direct us there. Publisher identity/business verification is separate from this technical eligibility question. Thank you, Jeff.

Answer 1

Hi jeff ostler Thank you for narrowing the initial test to retrieving previously authored samples and optionally saving a separately approved analysis, without new external model calls.

Three points require confirmation:

1. Supported Copilot experience: Can this operate within an existing consumer or Microsoft 365 Copilot conversation through a federated connector, or is a separate declarative/custom agent required?

2. Package and authentication: Which current Partner Center route, manifest/schema version, and authentication method apply? The published MCP certification and OAuth guidance appear inconsistent on Azure Key Vault, DCR, and whether intro.md is mandatory.

3. Execution and data handling: Does the intended host support asynchronous job status, bounded result retrieval, and reliable completion/truncation detection? Only explicitly approved inputs should be transferred, with separate permission required to save the analysis.

I will also consult our engineering team internally to assess technical feasibility and validate the proposed integration approach. Microsoft confirmation is still needed on the supported Copilot route and applicable certification requirements.

The initial test makes no new external model calls. We have private ChatGPT and Claude sample tests, but Microsoft compatibility and certification remain unverified. This is a presubmission eligibility enquiry, not a claim of existing support.

Thank you.