Microsoft Work IQ APIs Guide 2026: What Agent Builders Get on June 16

On June 16, 2026, Microsoft makes the Work IQ APIs generally available. If you are building agents that need to read, write, or reason over Microsoft 365 data — email, calendar, files, Teams, SharePoint, the org graph — this is the most important Microsoft release of the year. It quietly changes the cost and complexity of building production-grade agents on top of M365.

This guide unpacks what is in the GA release, how the API surface is shaped, what the new Copilot Credits billing model means in practice, and how Work IQ stacks up against Model Context Protocol, Anthropic’s Managed Agents, and the broader agent platforms covered in our Best AI Agent Platforms 2026 roundup.

The Short Version

Work IQ is Microsoft’s answer to the question “how do agents safely and cheaply interact with the entire M365 surface?” The GA release on June 16, 2026 ships three endpoint flavors (A2A, MCP, REST), collapses what used to be hundreds of per-data-type tools into roughly 10 generic ones, and introduces a consumption-based billing model decoupled from Copilot licensing.

• For agent developers: dramatically less integration code, faster agents, lower token bills.
• For enterprises: clearer billing through Copilot Credits, the same data boundaries Copilot already respects, and a real path off custom Graph API plumbing.
• For agent platform vendors: the M365 integration moat just shrunk.

If you are picking an agent stack for 2026 and your customers are on Microsoft 365, you have to evaluate Work IQ.

What Are the Work IQ APIs?

Work IQ is the semantic intelligence layer underneath Microsoft 365 Copilot. It continuously processes email, calendar, meetings, chats, files, and the org graph, and exposes that processed context to agents through a small set of generic tools.

Until now, that layer was effectively private — only Copilot could fully exploit it, while third-party agents had to glue together Graph API calls and figure out their own grounding and ranking. The GA release on June 16 opens the layer up to any developer building agents on Azure AI Foundry, Copilot Studio, or external platforms.

Three endpoint shapes ship at GA:

1. A2A (Agent-to-Agent) — for agents talking to other agents inside the M365 ecosystem.
2. Remote MCP server — a redesigned Model Context Protocol surface so any MCP-compatible client (Claude, ChatGPT, custom agents) can use Work IQ as a tool.
3. REST API — traditional HTTP endpoints for programmatic access.

All three sit on top of the same underlying Work IQ engine, so a query you ask through MCP and a query you ask through REST hit the same retrieval and ranking pipeline.

The 10-Tool Design (and Why It Matters)

The single most important design choice in Work IQ is the collapse to ~10 generic tools. Where the Graph API exposes hundreds of specific endpoints — list_emails, search_emails, send_email, read_calendar_event, create_calendar_event, list_files_in_folder, and so on — Work IQ exposes a small set of verbs with progressive disclosure.

In practice this means a single tool like search works across email, files, chats, and the org graph. The agent does not need to know in advance which data type holds the answer. It calls one tool, Work IQ does the routing.

The benefits for agent builders are concrete:

• Smaller tool list in the system prompt. Tool descriptions are one of the biggest token sinks in agent prompts. Going from 200+ tools to 10 saves serious context.
• Faster reasoning. With fewer tools, models pick the right one faster and call it correctly more often.
• Less integration code. You write against 10 well-documented tools instead of dozens of Graph API surfaces with their own auth quirks.
• Higher throughput. Microsoft claims faster response-tokens-per-second on agent loops using the consolidated tool surface.

If you have built an agent against the raw Graph API in the past two years, the porting effort is real — but the payoff is a system prompt that fits inside a reasonable token budget and an agent that actually reasons over the right data instead of fumbling tool selection.

What Each Endpoint Is For

A2A (Agent-to-Agent)

A2A is for agents within the M365 ecosystem to call other agents — for example, your custom Copilot Studio agent invoking a Sales-Copilot agent to enrich a CRM record. Auth is handled through the Microsoft Entra identity layer. The big use case here is multi-agent workflows where each agent owns a domain and you want to compose them without writing brittle orchestration code.

Remote MCP Server

This is the surface most external developers will use. Any MCP-compatible client — Claude Desktop, Claude Code, custom agents built on the OpenAI Responses API, Anthropic’s Claude Managed Agents — can connect to Work IQ as an MCP server and use the 10 generic tools to read and write M365 data.

If you have read our Model Context Protocol guide, this is the missing piece — Microsoft finally shipping a first-party MCP server that exposes the M365 surface to any agent. It is plausibly the biggest single MCP integration in the ecosystem.

REST API

For backend services, data pipelines, and use cases where you need synchronous, programmatic access without an agent loop. REST is also useful for custom orchestration where you want explicit control rather than letting the agent decide.

Pricing: Copilot Credits

Work IQ is billed through Copilot Credits, a new consumption-based currency. The key facts:

• Billing is independent of Microsoft 365 Copilot per-seat licensing. You can build agents on Work IQ without giving every user a Copilot license.
• Copilot Credits also cover Copilot Studio usage and other Microsoft AI services.
• Credits meter against tool calls and underlying compute, not raw token counts the way most LLM APIs work.
• Enterprise agreements roll Copilot Credits into the broader Microsoft commercial contract.

The practical effect for agent vendors: you can build a customer-facing agent that uses Work IQ under the hood, charge your customer per seat or per task, and have predictable underlying costs. This is a real shift from the old model where M365 access required a per-seat Copilot license on the customer side.

Cost Lever	Before Work IQ	After Work IQ
Per-seat Copilot license	Required for full M365 grounding	Not required for API access
Graph API plumbing	Custom code, custom auth, custom ranking	Microsoft-maintained
Tool count in prompt	50-200+ Graph endpoints	~10 Work IQ tools
Billing	Mixed Copilot + Graph + Azure	Unified Copilot Credits
MCP support	None official	First-party remote MCP server

Security, Compliance, and Data Boundaries

Work IQ respects the same data boundaries Microsoft 365 Copilot does. This is not a backdoor that bypasses tenant isolation or DLP policies. Three points matter for security teams:

1. Tenant isolation. A Work IQ call in tenant A cannot read data in tenant B.
2. Permission inheritance. The calling identity (user or service principal) sees only what they would see in Outlook, Teams, or SharePoint. There is no privilege escalation.
3. Audit logs. Every Work IQ tool call is logged in the M365 audit trail. Security teams can trace agent activity.

For broader context on agent-side risks — prompt injection, exfiltration through tools, governance — see our AI Coding Agent Security 2026 deep-dive and Best AI Agent Governance Tools 2026.

How Work IQ Compares to the Alternatives

Approach	Strength	Weakness	Best For
Work IQ APIs (June 16)	Unified M365 surface, MCP-native, consumption pricing	Microsoft ecosystem only	Agents whose users live in M365
Raw Microsoft Graph API	Maximum control, no extra licensing layer	Hundreds of endpoints, no built-in grounding	Existing Graph integrations you don’t want to migrate
Salesforce Agentforce	Deep Salesforce data, strong CRM workflows	Salesforce only	Salesforce-first orgs
Custom MCP servers	Full control, any data source	You build everything	Niche data sources
ChatGPT/Claude with web tools	Vendor-neutral, broad reach	No deep data access	Generic agents

If your users are split across M365 and Google Workspace, Work IQ doesn’t solve the Google side — see our Best AI Workflow Automation Tools 2026 and Google Antigravity 2 Review for that side of the stack.

How to Start Building on Work IQ

A pragmatic on-ramp for teams who want to ship something useful in the first month:

1. Pick a high-value workflow that lives in M365. Examples: meeting prep agent, inbox triage agent, internal-knowledge-base agent for SharePoint.
2. Stand up a Copilot Studio agent or a custom agent in Azure AI Foundry. Either gives you Work IQ access out of the box.
3. Wire the MCP endpoint if you are building outside the Microsoft stack. This is the path for Claude-based agents, custom Python agents, or anything not running on Azure.
4. Use the 10 generic tools first, before reaching for raw Graph. The whole point of Work IQ is that you do not need the long tail of Graph endpoints for 90% of agent tasks.
5. Instrument credit consumption from day one. Copilot Credits are predictable but easy to overspend if the agent loops on a poorly-bounded task.

Who Should Pay Attention

You absolutely need to evaluate Work IQ if:

• Your customers run on Microsoft 365 and your agent touches their work data.
• You are building a vertical agent (sales, HR, legal, finance) for M365 customers.
• You are currently maintaining a Graph API integration layer and it is bleeding engineering time.
• You are evaluating MCP-based architectures and want a credible enterprise data source.

You can probably defer if:

• Your customers run primarily on Google Workspace or Notion.
• Your agent does not need to read or write personal/org work data.
• You are still in early prototyping where Graph or hardcoded data fixtures are sufficient.

The Honest Verdict

Work IQ is the most consequential agent infrastructure release Microsoft has shipped since Copilot itself. The 10-tool design is the right architectural call, the MCP support is the right ecosystem call, and the Copilot Credits pricing is the right commercial call. The risk is the usual Microsoft risk — lock-in. You are building on top of a Microsoft-maintained ranking and retrieval pipeline. If Microsoft changes pricing, tool semantics, or product strategy, your agent’s behavior changes too.

For most teams building agents in 2026, that risk is acceptable given the alternative — owning the integration plumbing and grounding pipeline yourself, then losing every battle on cost and quality to whoever builds on Work IQ instead.

If you are evaluating where the agent ecosystem is heading more broadly, see our Best AI Agents 2026 roundup, Best AI Agent Frameworks 2026, and our deep dive on Microsoft Agent 365 — the consumer-facing Microsoft agent product that runs on the same Work IQ layer.

Frequently Asked Questions

Do I need Microsoft 365 Copilot licenses to use the Work IQ APIs? No. Work IQ API usage is billed through Copilot Credits separately from per-seat Copilot licensing. That said, the data your agent can reach is still scoped by the calling user’s M365 permissions.

Does Work IQ support write operations or only read? Both. The 10 generic tools include read, write, search, and act primitives. Writes go through the same permission and DLP checks the Copilot product uses.

Can Claude or ChatGPT agents call Work IQ? Yes, via the remote MCP server endpoint. Any MCP-compatible client can connect and use the tools.

How does pricing work for ISVs reselling agents built on Work IQ? Copilot Credits are charged to the tenant whose data is accessed. ISVs typically build customer-billed agents where the customer covers the Copilot Credits usage in their own M365 tenant.

Is this just rebranded Microsoft Graph? No. Work IQ sits above Graph and adds the semantic indexing, ranking, and grounding pipeline that powers Copilot. You can think of Graph as raw data access and Work IQ as the intelligence layer with consolidated tool surface.

When can I start building? The APIs are in preview today and reach GA on June 16, 2026. Production workloads should target the GA date; preview is fine for prototyping and design work.