Work IQ MCP tools
Want an answer? Use ask. Want the data? Use fetch for stored items, and call_function for data Microsoft Graph computes, such as calendarView or delta(). Don’t know where the data is? Use search_paths, then get_schema.
Each tool below shows whether the app’s harness gives it to the model, and the meter the app estimates it on; see Copilot Credits.
Answers from Copilot
Section titled “Answers from Copilot”Microsoft 365 Copilot, or a chosen agent, finds the relevant mail, chats, meetings and files, reasons over them and answers in prose with links. Best for open questions. Slower, and you see only the sources it cites.
What did Sarah say about the launch date?list_agents
Section titled “list_agents”Lists the agents available to the user; pass an agent’s ID to ask, for example to send an HR question to a published HR agent.
Exact data
Section titled “Exact data”Reads stored items by path: messages, events, files, people. Several paths per call, one result each. Best when your model or code needs complete, precise data, or the details of an item that ask pointed to. For remote MCP, fetch returns 25 items by default and 100 at most, and 10 chat messages per request; $skip and $skiptoken are blocked.
/me/messages?$filter=isRead eq false&$select=subject,fromcall_function
Section titled “call_function”Reads data that Microsoft Graph computes when you ask, one path per call: calendarView (meetings in a time range, with recurring ones expanded), delta() (what changed in mail, events, contacts or chat messages), /me/chats/getAllMessages(), /me/onlineMeetings/getAllTranscripts(…) and /sites/getAllSites(). In search_paths, most of them show as paths ending in (), but calendarView doesn’t; which ones exist depends on tenant policy.
/me/calendarView?startDateTime=…&endDateTime=…fetch_blob
Section titled “fetch_blob”Downloads a file’s bytes, such as a PDF, an Office file or a photo, optionally converted to PDF, so your own code can process the document: up to 4 MB for the raw file. The app withholds it because Base64 content would flood the model’s context.
Finding paths and fields
Section titled “Finding paths and fields”search_paths
Section titled “search_paths”Lists the available paths and their supported operations: the model’s first step when it doesn’t know where the data lives. For example, the filter calendar.
get_schema
Section titled “get_schema”Returns the fields and parameters of a path and operation: which fields to $select, or what a create or update requires.
Changes
Section titled “Changes”Changes are blocked by default by tenant policy; tenant admins can enable supported scenarios in that policy, so the current setting can differ by tenant. See Capabilities. A harness should ask the user before each change, and use get_schema first to learn the required fields. The app withholds all four tools.
| Tool | What it does | Example |
|---|---|---|
create_entity | Creates an item. | A meeting in /me/events, a draft in /me/messages |
update_entity | Changes an item. | Move a meeting to 4 pm, mark mail as read |
delete_entity | Deletes an item. | Delete a draft |
do_action | Runs an action with side effects. | Send mail, accept a meeting, forward, move to a folder |
The app estimates each of them as a Tools API call, at 0.1 credit.
Local CLI only
Section titled “Local CLI only”These depend on the CLI version. The app withholds them, and estimates no credits for them.
retrieve: raw search hits across Microsoft 365 and connected sources, with per-source metadata and citations. Useful for showing sources, but it isn’t in the public tool reference: treat it as a preview.get_debug_linkandaccept_eula: create a shareable debug link for a conversation, and accept the CLI’s license. Both need a person’s explicit confirmation.
Typical combinations
Section titled “Typical combinations”ask→fetch: find it with Copilot, then read the exact item it pointed to.search_paths→get_schema→fetch: when the model doesn’t know the path.fetchorcall_function→ your systems: combine exact Microsoft 365 data with your CRM, tickets or other tools.
The illustrations are AI-generated (MAI-Image-2.6). The screenshots and the demo show example data.