Published on August 12, 2026
Claude can now move beyond searching Microsoft 365 by drafting and organizing Outlook email, managing calendar events, and creating or updating files in OneDrive and SharePoint.
Anthropic expanded Claude’s Microsoft 365 connector on July 7, 2026, adding write capabilities that allow the AI assistant to take approved actions inside workplace accounts.
The connector was previously focused on retrieval. Claude could search Outlook messages, examine calendar information, locate documents in OneDrive or SharePoint, and review conversations in Microsoft Teams. With write tools enabled, it can now act on some of that information by preparing and sending email, organizing a mailbox, managing calendar events, and creating or updating cloud files.
This changes the connector from a search layer into a more active productivity tool. Instead of asking Claude to find information and then manually carrying the result into another Microsoft 365 application, users can ask it to continue into the next stage of the workflow.
The original value of the Microsoft 365 connector was its ability to retrieve workplace context without requiring users to upload every document or copy every email into Claude.
A professional could ask Claude to find a planning document in SharePoint, summarize a long Outlook thread, review meeting information, or locate a decision discussed in Teams. Claude could then analyze the retrieved material and produce a response in the conversation.
The new write tools extend that sequence. After examining an email thread, Claude can draft a reply or organize related messages. After reviewing people’s availability, it can create or update a calendar event. After analyzing project information, it can create a file in OneDrive or update an existing document stored in SharePoint.
The practical distinction is between assistance and execution. Search tools help a person understand what has happened. Write tools allow Claude to help move the work forward.
That may reduce one of the most repetitive parts of digital work: transferring information between applications. A user no longer needs to retrieve a message, summarize it separately, open Outlook, rewrite the summary as an email, and then manually create a follow-up meeting. Claude can potentially coordinate several of those actions from the same request.
Outlook receives some of the connector’s broadest new capabilities.
Claude can create, update and delete drafts, including reply and reply-all drafts. It can send or forward messages, move conversations, apply categories, organize multiple emails, configure automatic replies, and create or remove inbox rules.
This could support workflows such as reviewing a group of project emails, identifying which messages require attention, categorizing them by topic and preparing responses for review. A manager returning from leave could ask Claude to summarize important conversations, organize them and draft follow-ups without manually opening every thread.
Claude can also send email when the relevant permissions are enabled. However, the ability to generate a polished message does not guarantee that its interpretation of a conversation is correct. External communications involving clients, legal matters, sensitive information or binding commitments still require careful human review.
Anthropic has added several restrictions. Emails sent by Claude include a technical attribution header identifying them as agent-initiated. Write tools are subject to limits on actions, recipients and sends, and attachments are not currently supported when Claude drafts, forwards or sends email through the connector.
The connector can now create, update and delete calendar events. It can also respond to invitations by accepting, declining or tentatively accepting them.
This makes it possible to connect scheduling with information Claude has already retrieved. After summarizing a discussion about an unresolved project issue, Claude could identify the people involved, examine availability and prepare a follow-up meeting.
For professionals who coordinate across several teams, this may reduce the administrative work involved in translating conversations into scheduled actions. The benefit is not simply that AI can create a meeting. It is that the meeting can be informed by the emails, documents and calendar context surrounding the underlying task.
Calendar actions can still carry consequences. An incorrect time, participant list or description may create confusion, while deleting or changing an existing event can affect several people. Anthropic therefore does not support a permanent “always allow” setting for creating or updating calendar events. The user retains an approval layer for these more consequential actions.
Claude can also create and update files in OneDrive and SharePoint, provided the user already has permission to work with the relevant location.
The underlying tools can upload new files, replace existing content, create folders, rename or move items, copy files, and send deleted items to the recycle bin. This allows Claude to participate more directly in document-based workflows rather than only analyzing files and returning its work inside a chat.
A project team could ask Claude to consolidate information from several SharePoint documents and save an updated report in the appropriate folder. An operations professional might request a revised procedure based on recent emails and meeting notes. A student or individual user with an eligible work account could organize research materials and save a generated document directly to OneDrive.
These capabilities do not give Claude unrestricted access to the organization’s storage. The connector uses delegated permissions, meaning Claude acts on behalf of the individual user. It can reach and modify only content that the person is already allowed to access through Microsoft 365.
The Microsoft 365 connector does not offer the same level of action across every application.
Claude can search Microsoft Teams chats and channel discussions that the user is permitted to access. It can use those conversations as context for summaries, decisions and other work. However, it cannot post Teams messages, modify Teams settings or change channel permissions through the connector.
This creates an intentional boundary. Teams can inform a workflow, but Claude cannot currently complete that workflow by speaking inside the conversation.
For example, Claude could review a Teams discussion, identify the agreed next steps, create a supporting document in SharePoint and schedule a follow-up meeting. It could not post the finished summary back into the Teams channel through this connector.
The limitation also illustrates how enterprise AI access is likely to develop unevenly. Some services may support extensive write actions, while others remain read-only because of technical, administrative or safety considerations.
Write capabilities are not activated automatically simply because a user connects Microsoft 365.
The administrator setup process requires a Microsoft Entra Global Administrator to approve additional Microsoft Graph permissions. The organization must then enable write tools in Claude. Companies that connected Microsoft 365 before the write-tool launch have them blocked by default until an administrator deliberately activates the new permissions.
Enterprise organizations can limit write access to selected users through custom roles. Administrators may also revoke particular Microsoft permissions, while individual members can turn off specific tools in their Claude connector settings.
This layered approach separates three decisions: whether Claude may connect to the organization, which Microsoft 365 resources it may access, and whether it may modify those resources.
The connector is listed as available across Free, Pro, Max, Team and Enterprise Claude plans. However, it requires a Microsoft 365 work account connected to a Microsoft Entra tenant. Personal addresses such as Outlook.com, Hotmail or Live accounts are not supported. Team and Enterprise organizations also require their Claude owner to enable the connector before members can use it.
The connector’s expansion reflects a broader shift in AI productivity. Assistants are moving from retrieving and summarizing information toward changing the systems where work is stored.
That progression can save time, but it also makes mistakes more consequential. An inaccurate summary is a content problem. An incorrectly sent email, deleted calendar event or overwritten file is an operational problem.
Organizations therefore need to decide which activities can be delegated safely. Drafting an internal email for review carries less risk than sending a message to a client. Creating a new report is more reversible than replacing a critical file. Searching a calendar is less consequential than cancelling a meeting.
Anthropic’s safeguards—including delegated permissions, administrative controls, action limits and additional approval requirements—provide boundaries, but they do not remove the need for human judgment. Teams should match Claude’s access to the risk and reversibility of each workflow.
Claude’s Microsoft 365 connector is becoming more than a workplace search tool. By adding write capabilities, Anthropic is allowing Claude to continue from understanding information into performing selected actions across Outlook, calendars, OneDrive and SharePoint.
The most useful workflows may be those that connect several stages of ordinary knowledge work: finding the relevant context, interpreting it, preparing a response, organizing the follow-up and saving the result in the correct location.
The expansion also makes governance more important. Claude can act only within existing user permissions and administrator-approved scopes, but those actions may still affect colleagues, customers and shared company information.
The wider significance is that AI productivity is moving beyond helping people decide what to do. Tools such as Claude are increasingly being designed to carry approved decisions into the applications where work actually happens.
Anthropic Help Center — Claude Release Notes — Microsoft 365 write tools announced July 7, 2026.
Anthropic Help Center — Connect to Microsoft 365 — Updated July 2026; accessed August 5, 2026.
Anthropic Help Center — Set Up the Microsoft 365 Connector — Updated July 7, 2026; accessed August 5, 2026.
Anthropic Help Center — Microsoft 365 Connector Security Guide — Updated July 7, 2026; accessed August 5, 2026.
Anthropic Help Center — Use Connectors to Extend Claude’s Capabilities — Updated July 2026; accessed August 5, 2026.
