Connectors give Claude permission-scoped access to the tools each team already uses. Here is how they work, how admins roll them out and three handoffs they fix.

Table of Contents

  1. Why Enterprise Work Stalls Between Teams
  2. What Claude Connectors Are
  3. How Connectors Work Under The Hood
  4. Shared Setup For Teams And Enterprises
  5. Example 1: From Sprint Board To Release Page
  6. Example 2: One Answer From Many Microsoft 365 Sources
  7. Example 3: From Customer Complaint To Engineering Ticket
  8. Pulling Live Context From The Web
  9. Connectors Beyond The Chat Window
  10. Guardrails To Set On Day One
  11. Where To Start
  12. References

Why Enterprise Work Stalls Between Teams

Most enterprise work does not fail inside one team. It fails at the handoff. A salesperson logs a customer request in the CRM, a product manager rewrites it as a ticket, an engineer reads a thinner version of it and support hears about the fix weeks later.

Each handoff loses context because each team lives in a different tool. An AI assistant that only reads what someone pastes into a chat window inherits the same gap. Claude connectors close it by letting Claude work inside the tools where the record already sits.

Four teams sharing context through Claude and connectors Claude plus connectors Sales CRM records Product Jira or Linear tickets Engineering GitHub and Sentry Support Intercom and Slack old handoff: copy, paste, summarize
Figure 1. Teams keep their own tools. Claude reads and writes in each one within every user's existing permissions.

What Claude Connectors Are

Connectors let Claude reach your apps, retrieve your data and take actions inside connected services, as described in Anthropic's guide to using connectors. You can ask Claude to create a Linear issue, send a Slack message or search Google Drive from one conversation.

The same connectors work across Claude on the web, Claude Desktop, Claude Code and the API through the MCP Connector in the Claude API. That matters for mixed teams. An analyst in the browser and a developer in a terminal reach the same systems through the same integration.

Anthropic keeps a public directory of Claude connectors where each entry lists use cases, read and write capabilities and availability. Organizations can also add custom connectors for internal systems that will never appear in a public catalog.

How Connectors Work Under The Hood

Every connector runs on the Model Context Protocol (MCP). Anthropic introduced the Model Context Protocol in November 2024 as an open standard for linking AI assistants to the places where data lives, such as business tools, content repositories and development environments.

MCP solves an integration math problem. Without a shared protocol, every AI app needs a custom bridge to every data source. With MCP, each tool exposes one server and each AI app speaks one client protocol.

Point-to-point integrations compared with MCP Without A Shared Protocol With MCP Chat app IDE agent Custom bot CRM Docs Tickets Email 3 apps x 4 tools = 12 custom bridges Chat app IDE agent Custom bot MCP CRM Docs Tickets Email one client protocol, one server per tool
Figure 2. A shared protocol replaces a grid of one-off integrations.

The protocol now reaches well beyond Anthropic. The Wikipedia entry on Model Context Protocol records that Anthropic donated MCP to the Agentic AI Foundation under the Linux Foundation, and reports more than 10,000 MCP servers in production by mid-2026.

Remote connectors run from Anthropic's cloud, not from your laptop. The guide to custom connectors using remote MCP notes this holds even for Cowork and Claude Desktop, so a server behind a corporate firewall needs a public endpoint or an allowlist for Anthropic's IP ranges.

Request path of a remote Claude connector Employee web, desktop, Code Claude Anthropic cloud Remote MCP server OAuth sign-in Business app Slack, Drive, Jira Permission check at the source Claude sees only what this employee can already open in the app
Figure 3. The request path for a remote connector, with access set by the source system.

The most important design choice sits in the permission model. Claude inherits each person's permissions from the connected service. If someone cannot open a file, channel or record in the source system, Claude cannot reach it on their behalf either.

Shared Setup For Teams And Enterprises

On Team and Enterprise plans, setup splits into two steps. An Owner or Primary Owner enables a connector for the organization, and then each member signs in with their own account before using it.

This split is what lets work carry across people. Governance owners configure the connector once, while access still follows each employee's identity in the source system. An Enterprise-managed auth option, currently in beta, lets admins authorize a connector once so members inherit access on first login.

Four-step connector rollout on Team and Enterprise plans 1 Owner adds connector Organization settings 2 Owner sets tool rules allow, approve, block 3 Member signs in own account or org auth 4 Work starts in chat toggle per conversation
Figure 4. How a connector reaches every member of an organization.

Owners can also restrict what a connector may do across the whole organization. Each tool can be set to Always allow, Needs approval or Blocked, so a team might let Claude search email without sending it. Members cannot override these organization-wide limits.

Members who need a missing tool can request it from the directory, and Owners approve or decline it from the admin notifications tab. Requests are on by default for Team plans and off by default for Enterprise plans.

Example 1: From Sprint Board To Release Page

Product to everyone

A product manager closes a sprint in Jira. Instead of copying tickets into a document, they ask Claude to summarize the completed work and draft the release page in Confluence. In its announcement of Claude Integrations for remote MCP, Anthropic said the Atlassian connector lets Claude summarize and create multiple Confluence pages and Jira work items at once.

The handoff gain is accuracy. Marketing, support and sales read a page built from the actual tickets instead of someone's memory of the sprint, and engineers keep the afternoon they would have spent writing it.

Example 2: One Answer From Many Microsoft 365 Sources

HR and operations

A new manager asks about the remote work policy. The answer is spread across an HR document in SharePoint, a clarifying thread in Outlook and a discussion in Teams. Anthropic's Microsoft 365 connector and enterprise search announcement uses this exact case, with Claude pulling from all three sources into one report.

The security model suits regulated teams. According to the Microsoft 365 connector security documentation, the connector uses delegated permissions, never stores passwords and leaves files in the customer's tenant. Admins can revoke single Entra permissions, for example blocking SharePoint while leaving Outlook open.

Example 3: From Customer Complaint To Engineering Ticket

Support to engineering

A customer reports a checkout bug in an Intercom chat. A support lead asks Claude to read the conversation, check related Slack threads and file a bug in Linear with the reproduction steps attached. Coverage of the launch by Analytics India Magazine on Claude Integrations described this Intercom bug-reporting flow.

The Claude and Slack connector overview adds that Claude searches only the channels a person can already access. The engineer receives a ticket with the customer's own words and the internal discussion in one place.

Pulling Live Context From The Web

Connectors cover internal systems, and web tools cover the rest. When Anthropic launched Integrations it also extended Research so that Claude could search the web, Google Workspace and any connected app in one task.

Developers get the same pattern in the API. The Claude web fetch tool documentation explains that it retrieves full web pages and PDFs, and that Claude can only fetch URLs supplied by the user or returned by earlier search and fetch results. That rule limits where data can travel.

Research flow combining web and connector sources Web search finds public sources Web fetch reads full pages and PDFs Connected apps internal records via MCP Claude compares and checks Cited deliverable brief, page or ticket
Figure 5. Public and internal sources meet in one task and leave as one sourced output.

Connectors Beyond The Chat Window

Employees do not have to name a connector every time. Once an app is connected, Claude can bring it into a conversation on its own when it fits the request, so a question about last week's open bugs reaches the tracker without a menu hunt.

Some connectors go further than text. Entries with the Interactive badge in the directory render live interfaces such as dashboards, task boards and design tools inside the conversation, so a reviewer can work with the real board instead of a description of it.

Developers reach the same layer from code. The API's MCP Connector lets an internal agent call the same remote servers that employees use in the Claude app, which keeps one integration behind both the human workflow and the automated one.

Guardrails To Set On Day One

Connectors give Claude real reach into business systems, so the rollout plan matters as much as the tool list. These controls come from Anthropic's own connector documentation:

  • Start with read access: Set write and delete tools to Needs approval until teams trust the output.
  • Vet custom connectors: Anthropic warns that custom connectors can reach services it has not verified, so connect only servers from trusted organizations.
  • Watch shared credentials: A custom connector that uses a shared credential reaches everything that credential can access, not just what each user can see.
  • Protect your domain: Enterprise Owners can stop services on their verified domains from being connected to Claude accounts outside the organization.
  • Load tools on demand: With ten or more active connectors, the On demand tool access mode leaves more room in each conversation.

Plan around one limit as well. On Team and Enterprise plans, connectors work only in private projects, and chats with synced content cannot be shared. The output Claude writes into Jira, Confluence or Linear becomes the shared record, and that record is what the next person picks up.

Where To Start

Pick the single handoff that costs your organization the most rework, such as sales to product or support to engineering. Connect the tools on both sides of it, set write actions to Needs approval and run it for two weeks.

Measure how often the receiving team has to ask a follow-up question. If that number drops, add the next handoff. Connectors pay off one boundary at a time.

References

  1. Introducing the Model Context Protocol, Anthropic, November 2024.
  2. Claude Can Now Connect To Your World, Anthropic.
  3. Use Connectors To Extend Claude's Capabilities, Claude Help Center.
  4. Get Started With Custom Connectors Using Remote MCP, Claude Help Center.
  5. Connectors Directory Documentation, Claude Docs.
  6. Claude Connectors Directory, Claude.
  7. MCP Connector, Claude Platform Docs.
  8. Claude And Your Productivity Platforms, Anthropic, October 2025.
  9. Microsoft 365 Connector Security Guide, Claude Help Center.
  10. Claude And Slack, Claude.
  11. Web Fetch Tool, Claude Platform Docs.
  12. Model Context Protocol, Wikipedia.
  13. Claude Users Can Now Connect Apps And Run Deep Research Across Platforms, Analytics India Magazine.