Claude is an AI assistant made by Anthropic, which calls itself an AI safety and research company and is set up as a public benefit corporation. Staff use Claude in a browser or in its desktop and mobile apps, and developers build it into other software through an API.
In May 2026, Anthropic said its run-rate revenue had crossed $47 billion, in the same announcement as a funding round that valued the company at $965 billion.
What staff hand to Claude
Claude works inside Microsoft Office as well as in its own apps. With Claude for Excel, a finance manager can ask what any cell is doing, or change an assumption without breaking the formulas that depend on it. The Office add-ins are on every paid plan. Edits show up as highlighted cells in Excel and tracked changes in Word, and Anthropic says nothing goes out or gets saved until the user says so.
Longer jobs go to Cowork, which Anthropic now folds into Claude itself. The user picks the folders and tools, and Claude works through a multi-step task inside them. A task can run on a schedule while nobody watches, and deleting anything needs the user’s approval.
Claude’s connectors directory
Anthropic’s connectors directory lists more than 900 connectors, filed under categories like commerce and shopping, financial services, productivity, and sales and marketing. Many are built by the software company itself and carry an “Anthropic verified” badge.
What a connector can do comes down to its tools. The Xero connector, made by Xero, has read-only access. Claude can pull profit and loss or aged receivables, so a question about overdue invoices gets an answer and the invoices themselves stay as they were.
The Shopify connector, made by Shopify, goes further. Besides looking up orders and customer details, it can edit products and create discount codes, which means a request typed in chat can change what’s live in the store.
The Microsoft 365 connector searches Outlook, SharePoint, OneDrive and Teams, on every plan from Free to Enterprise. Its permissions start read-only. An organization admin can switch on write tools, and then Claude can send email, manage calendar events, edit files and post Teams messages on a member’s behalf. A Microsoft Entra Global Administrator has to approve the connector for the tenant first. Claude mirrors each person’s existing Microsoft 365 permissions, so nobody reaches a file through Claude that they couldn’t open already.
Connectors and MCP, underneath
Every connector is an MCP server. Anthropic introduced the Model Context Protocol in November 2024 as an open standard for connecting AI assistants to the systems where data lives, in place of a custom integration for every new data source.
In December 2025, Anthropic handed MCP to the Agentic AI Foundation, a fund under the Linux Foundation that it co-founded with Block and OpenAI. By then ChatGPT, Cursor, Gemini, Microsoft Copilot and Visual Studio Code had all adopted it.
So the difference between a connector and “an MCP server” is mostly who set it up. A directory connector is an MCP server a vendor published and Anthropic listed. A custom connector is any remote MCP server added by its web address, which is how a system with no directory listing gets connected.
Desktop extensions run on the computer itself, for tools that need local folders or a database on that machine, and they only work in Claude Desktop and Claude Code.
Owner controls on Team and Enterprise
On Team and Enterprise plans, an Owner switches each connector on for the organization. Enabling one makes it available without granting access, and each person still signs in to the underlying service with their own account. Owners can then set a connector’s tools, grouped into types like read-only and write or delete, to Always allow, Needs approval or Blocked for everyone in the organization.
Both plans include single sign-on and admin controls for remote and local connectors. Enterprise adds SCIM provisioning, audit logs, role-based access, custom data retention and a Compliance API.
For its commercial products, including Team and Enterprise plans and the API, Anthropic’s policy is that it won’t use inputs or outputs to train its models by default. Feedback is the exception: rating a response with a thumbs up or down shares that conversation with Anthropic, which may use it in training. Owners can switch the rating buttons off for the whole organization in its Data and Privacy settings.
Building Claude into a workflow
The Claude API runs Claude inside a business’s own software instead of a chat window. A script can hand Claude a document or a batch of customer emails and act on the structured answer that comes back.
Developers create API keys in the Claude Console, where workspaces keep projects apart and control spending for each one. Official SDKs cover Python, TypeScript, C#, Go, Java, PHP and Ruby. Claude is also sold through the large cloud platforms, including Amazon Bedrock and Microsoft Foundry, for businesses that would rather pay through an existing cloud account.
Claude and the rest of the stack
Claude’s connectors and API give a business plenty of ways to put it to work. A finance manager can ask Xero and Excel the same question in one chat, a support team can have Claude draft replies from a customer’s order history, and a script can hand Claude a batch of supplier PDFs and act on the line items that come back.
Every answer a connector gives comes from the data in the connected system, duplicate customers and unmatched payouts included. What AI-ready means for a 20-person company covers the groundwork, and a readiness scorecard puts a number on where a business stands.