Can AI Agents Access Your Microsoft 365 Data? Work IQ APIs Go GA With A2A and MCP

核心结论

If you're searching "how do AI agents access Microsoft 365 data with real context," the short answer is: Work IQ APIs reached general availability on June 16, 2026. This isn't another chatbot wrapper—it's Microsoft's production-grade intelligence layer that exposes the same reasoning engine Copilot uses to any agent you build, through three protocols: Agent-to-Agent (A2A), Model Context Protocol (MCP), and REST. You pay per use, and you don't need a Copilot license to start.

What Microsoft Work IQ Actually Is

Microsoft Work IQ is the intelligence layer behind Microsoft 365 Copilot—the system that understands your email, calendar, documents, Teams messages, and organizational relationships, then reasons about them. Before June 16, only Copilot itself could access this layer. Now, Microsoft has opened it as a set of APIs available to any developer building AI agents.

Announced at Microsoft Build 2026 on June 2, the Work IQ APIs give external agents governed access to Microsoft 365 context. This means an agent you build—whether it runs in Claude Code, n8n, a custom Python script, or any other runtime—can now ask questions like "what's the status of the Q3 budget review?" and get back a cited, reasoned answer based on actual data from your organization's Microsoft 365 environment.

The key word here is "governed." Work IQ doesn't hand your agent raw Microsoft Graph API access and say "good luck." Every request respects the permissions of the authenticated identity. An agent can only see what that user or service principal is allowed to see. There's no backdoor to org-wide data.

Three Protocols, One Intelligence Layer

Work IQ doesn't lock you into a single integration pattern. Microsoft ships three protocol options, and you pick the one that fits your agent architecture:

Agent-to-Agent (A2A) is Google's open protocol for agent interoperability. If your agent speaks A2A, it can discover Work IQ as a remote agent and delegate tasks to it. Your agent treats Work IQ like another team member it can hand work to—"go check the status of the Q3 budget review and come back with a summary."

Model Context Protocol (MCP) is Anthropic's open standard for connecting LLMs to external tools. Work IQ exposes a hosted MCP server with standard tools for interacting with Microsoft 365 data. If your agent already supports MCP—as Claude Code, Hermes Agent, Cursor, and a growing list of tools do—integrating Work IQ is mostly plug-and-play. The MCP server handles authentication, rate limiting, and data governance, so your agent doesn't need to understand Microsoft Graph API internals.

REST API is the traditional HTTP approach. Microsoft provides REST endpoints for the Chat API and direct data access. This is the fallback for agents that don't speak A2A or MCP, or for developers who want maximum control over request formatting.

The Chat API: Copilot-Grade Reasoning Without Copilot

The most significant endpoint is the Chat API. When your agent sends a natural language question, Work IQ doesn't just run a keyword search against Microsoft 365 data. It performs the same multi-step reasoning Copilot does: retrieving relevant emails, documents, and calendar entries, cross-referencing them, and synthesizing a cited answer.

Each response includes citations pointing back to the source documents, emails, or meetings that informed the answer. This makes the output auditable—a critical requirement for enterprise deployments where hallucinated facts can have real consequences.

The Chat API also returns responses in the same format a user would get from Copilot: a natural language answer with inline citations, not a JSON blob of raw search results. This means your agent can present the answer directly to end users without additional formatting work.

Pricing: Consumption-Based, No Copilot License Required

Work IQ API usage is independent of Microsoft 365 Copilot licensing. You don't need to buy Copilot seats to use the APIs. Instead, pricing is consumption-based: you pay per API call or per token processed.

Microsoft hasn't published exact per-call rates yet, but the Build 2026 announcement confirmed the consumption model. This is significant for two reasons. First, it lowers the barrier to experimentation—a solo developer can prototype an agent without committing to enterprise Copilot licenses. Second, it means enterprises can budget Work IQ separately from their Copilot rollout, treating it as infrastructure rather than a per-user productivity tool.

What This Changes for Agent Builders

Before Work IQ, building an agent that understood organizational Microsoft 365 data meant wrestling with Microsoft Graph API—a sprawling set of REST endpoints that give you raw data (emails as JSON, calendar entries as objects) but no reasoning. Your agent had to fetch the data, feed it to an LLM, and synthesize answers itself.

Work IQ shifts that burden to Microsoft's infrastructure. Instead of fetching 50 emails and asking an LLM to summarize them, your agent sends one question to the Chat API and gets back a reasoned, cited answer. This reduces token costs, simplifies agent architecture, and—crucially—keeps sensitive organizational data inside Microsoft's governed environment rather than sending raw email contents to a third-party LLM provider.

For agent platforms like n8n, the MCP integration is particularly interesting. An n8n workflow can include an "Ask Work IQ" node that queries organizational context without leaving the governed Microsoft 365 boundary. The same applies to any MCP-compatible agent runtime—Claude Code agents, Hermes Agent workflows, or custom Python agents built on the MCP SDK.

The Competitive Landscape

Work IQ enters a market where "enterprise context for AI agents" is becoming a distinct product category. Google's equivalent lives inside Gemini for Workspace and the Antigravity agent platform. Slack has its own AI-native search layer. Anthropic's Claude Enterprise offers organization-aware features.

Microsoft's advantage is the data gravity of Microsoft 365 itself. With over 400 million commercial seats, the volume of organizational context already inside the Microsoft ecosystem dwarfs competitors. Work IQ makes that context accessible regardless of which AI model or agent framework you use—you're not locked into Microsoft's Copilot or Azure OpenAI models.

The A2A protocol support is strategically notable. By adopting Google's agent interoperability standard rather than building a proprietary equivalent, Microsoft is betting the agent ecosystem will be multi-vendor. An agent built on Claude can talk to Work IQ via A2A just as easily as one built on Gemini or GPT models.

What to Watch

The GA release is the starting line. Several items on the roadmap deserve attention:

Windows-level agent runtime: Microsoft has signaled a deeper agent runtime coming in a later 2026 Windows update. This would let agents run with system-level context—understanding which applications are open and what the user is working on.

Pricing transparency: The exact consumption rates will determine whether Work IQ becomes a default integration or a premium feature reserved for high-value workflows.

Third-party MCP tools: The hosted MCP server currently exposes Microsoft-built tools. Whether third-party developers can publish their own tools to the Work IQ MCP server will determine ecosystem extensibility.

Enterprise governance: As agents gain access to organizational context, IT teams will need fine-grained controls over which agents can ask which questions. Microsoft's Entra ID and Purview infrastructure provides a foundation, but agent-specific policy controls are still evolving.

Bottom Line

Microsoft Work IQ APIs reaching GA on June 16, 2026 isn't just another API launch. It's Microsoft opening the intelligence layer that powers Copilot to the broader agent ecosystem. For developers building AI agents that need organizational context, this eliminates the "build your own reasoning pipeline on top of raw Graph API data" step that has been standard for years. The three-protocol design—A2A, MCP, REST—shows Microsoft is serious about meeting developers where they are, not where Microsoft wishes they were.