The roadmap lists the feature under Work IQ and Microsoft Copilot (Microsoft 365), on the Developer platform, for the Worldwide (Standard Multi-Tenant) cloud. Its status is "In development." The estimated dates are November 2026 for preview and January 2027 for general availability. Microsoft's own announcement material, however, describes the feature as already generally available. That conflict matters for anyone planning a rollout, and it's covered below.
What Work IQ MCP is, and why one switch isn't enough
Work IQ MCP is Microsoft's Model Context Protocol server for Microsoft 365. Through one endpoint, AI agents can read, create, update and delete Microsoft 365 items, call Microsoft 365 Copilot, retrieve grounding content and look up API schemas.
Microsoft built it around a small set of generic tools that act on resource paths. It does not create a separate tool for every workload. Microsoft's Learn documentation gives three examples:
fetch /me/messagesreads mailcreate_entity /me/eventscreates a calendar eventdo_action /me/sendMailsends mail
The tool names are the verbs, and the resource paths say what they act on. The documented tools are:
| Category | Tools |
|---|---|
| Entity tools | fetch, fetch_blob, create_entity, update_entity, delete_entity, do_action, call_function |
| Copilot tools | ask, retrieve (public preview), list_agents |
| Schema tools | get_schema, search_paths |
This design keeps the list of tools short, but it makes governance harder. If one do_action tool can either send an email or do something harmless, the useful control point is the path and the operation, not the tool name.
Summary: Work IQ MCP uses about a dozen generic tools to reach a large part of Microsoft 365, so access controls need to work at the level of individual operations.
Today's baseline: governance at tenant level
Microsoft's Learn page on Work IQ MCP policy governance describes how things currently work. In the initial release, Work IQ MCP provides tenant-level policy control. In the Microsoft 365 admin center, tenant administrators can turn Work IQ MCP policy on or off for the tenant. The same page says per-user, per-app, per-agent and scenario-specific policy templates are not part of that initial control surface. It also says more granular controls might be added later, which is what roadmap 570671 now commits to.
The policy engine is based on Rego, the policy language from Open Policy Agent. According to Microsoft, it evaluates these inputs on each request:
- the resource path
- the operation
- the request content, where applicable
- user and tenant context
For Work IQ MCP, policy governance doesn't replace Microsoft Entra authentication, OAuth permissions, or the signed-in user's existing Microsoft 365 permissions. It is an extra check on top of them. If policy allows a request, the user still can't reach a resource they don't already have permission for. If policy denies a request, it stays denied even when the user and app are fully authorised.
An analysis by the Swedish consultancy Tyto adds a useful point. Microsoft's design principles mention four broad OAuth permissions, but Tyto found that the permission reference lists only one, WorkIQAgent.Ask, which grants read and write access. Its conclusion: "The block on writing therefore does not sit in the permission, it sits in the tenant policy." That is why the policy layer carries most of the weight here.
Under the current model, mutation operations are blocked by default. That covers create, update, delete and actions that change data, such as sending email. Administrators can turn on supported mutation scenarios when they need them. Integration guide author Can Dede notes that when policy rejects a request, the server usually returns a policy-denied error (usually a 403 Forbidden).
Summary: Today's control is one tenant-wide policy that blocks writes by default. Because the only listed OAuth permission is broad, the policy is where most of the real control happens.
What "granular" adds
The roadmap names three ways to restrict access: trusted platforms, workloads and supported operations. In practice, that could separate two decisions that are currently bundled together:
- Which integrations can connect. For example, an approved internal agent platform could be allowed while an unknown client is not.
- What those integrations can do. For example, reading calendars could be allowed while sending mail is blocked.
Microsoft's Azure blog post introducing the new Microsoft IQ describes the feature in slightly different terms: "Granular MCP policies (generally available) scope access by path, audience, and trusted app rather than relying on a tenant-wide switch." The two descriptions are similar but not identical. "Audience" and "trusted app" don't map exactly onto "platforms" and "workloads," and neither source defines the terms.
Several details are not yet public:
- which platforms or apps will count as "trusted"
- which workloads and operations are covered
- how administrators will configure the new rules
- licensing requirements
- whether the granular rules will be on by default
The current blocked-by-default rule for writes belongs to the tenant-level policy. Nothing confirms that the new granular controls will use the same default.
The date conflict
Microsoft's roadmap still marks the item "In development," with preview in November 2026 and general availability in January 2027. The Azure blog post, published around Microsoft Ignite 2026, labels granular MCP policies as generally available. WindowsForum's own Ignite coverage also lists them among the GA Work IQ features, alongside long-running tasks and agent-to-agent support.
There are several possible explanations. The roadmap may be slow to update, the rollout may be staged, or the GA label in the announcement may refer to something narrower than the roadmap item. None of the available evidence settles it. The practical advice is to rely on what your own Microsoft 365 admin center shows, not on either label.
Also keep this separate from agent policy governance with human approval, which pauses an agent for sign-off before consequential actions. The Azure blog lists that feature as public preview. Granular MCP policies decide whether a request is allowed. Human approval adds someone to sign off. They are different controls.
What admins and developers should do now
Administrators
- In the Microsoft 365 admin center, go to Agents > Tools, select Work IQ MCP, and open the Policy tab when it's available. That's where the current tenant-level setting lives, and it's the most likely place for granular options to appear, though Microsoft hasn't confirmed this.
- Allow up to 24 hours for saved policy changes to apply across the tenant.
- If Work IQ has never been used in your tenant, policy configuration can fail because the Work IQ service principal doesn't exist yet. Microsoft documents how to provision it in its tenant-enablement steps.
- If the policy settings don't appear, check that your tenant has access to Work IQ MCP and that your admin account has the required permissions.
- Start listing which agent platforms you trust and which write actions each one actually needs. That list will be useful whatever the final configuration screens look like.
Developers
- Don't assume that a path or operation allowed in one tenant is allowed in another.
- Treat policy denials as governance decisions, not temporary errors, so don't retry them automatically.
- Show policy denials clearly so users and admins know governance blocked the action rather than a bug.
Granular policies will make these practices more important, because whether a request succeeds will depend on which client sent it as well as on the tenant.
Analysis
The direction is right, and somewhat overdue. A design with a handful of generic tools means that a coarse on/off switch for writes is too blunt. You either block everything useful or you allow sendMail for every connected agent. Restricting access by trusted platform and by operation is how least privilege should work for agents acting inside Microsoft 365.
There are open questions. Who decides what counts as a "trusted platform," and can an MCP client misrepresent what it is? The roadmap doesn't say. Until Microsoft publishes the matching Learn documentation, admins should treat the new controls as an additional layer and keep relying on Entra Conditional Access, consent policies and existing user permissions as their main protection.
Bottom line: Work IQ MCP is moving from a single tenant-wide policy to controls over who can connect and what they can do. Microsoft's own sources disagree on whether that is in preview or generally available, so check your admin center and expect the details to change.
References
- Work IQ: Granular MCP policies Microsoft 365 Roadmap · 2026-11-17T19:53:11.795350Z
- Introducing the new Microsoft IQ: Putting your organization's intelligence to work in every agent azure.microsoft.com
- Microsoft Copilot Roadmap | Microsoft 365 microsoft.com