Best Claude Connectors in 2026: Which to Add and How

Claude connectors let Claude read from and act in the apps your team already uses, such as Google Drive, Slack, Jira and Salesforce, while respecting each person's existing permissions. The best Claude connectors for a work team are the ones that cover where decisions actually live: your files, your chat, your tickets and your CRM.

Anthropic's Connectors Directory listed 974 connectors when I checked it on October 19, 2026, so the hard part is not finding one. It is picking the few that earn their place, setting them up so admins stay in control, and knowing what they cannot do. This guide covers which connectors to add, how to add them, and where they fall short. If you want the protocol side (claude mcp add, scopes, the API), that lives in my Claude MCP guide.

Every connector fact below comes from Anthropic's help center, Anthropic's docs, or the connector's own listing in the directory, read on October 19, 2026.

What are Claude connectors?

A Claude connector is an MCP server that Claude can call to fetch data or take an action in another app. MCP, the Model Context Protocol, is an open standard, which is why the same server can often be used from other AI clients too. Per Anthropic's connector guide, "Claude inherits each person's permissions from the connected service." If you cannot open a file or channel in the source system, Claude cannot reach it through the connector either.

Three things are worth knowing up front:

• One catalog, many surfaces. The same directory serves claude.ai, Claude Desktop, mobile, Cowork and Claude Code, per Anthropic's directory docs. Connectors also work through the API via the MCP connector.
• Labels describe review, not safety. Directory entries are marked Verified (Anthropic tested the tools), Community (screened, not reviewed in depth) or Custom (added by you). Anthropic's verification page says plainly that verification "isn't a security audit."
• Most are built by the vendor. Google built the Drive and Gmail connectors, Slack built Slack's, Atlassian built the Rovo MCP server. Anthropic built the Microsoft 365 connector. The developer controls which tools a connector exposes and can change them after review.

The best Claude connectors at a glance

Sources: Anthropic's Google Workspace guide, Microsoft 365 docs, Slack docs and each connector's directory listing, checked October 19, 2026.

Which Claude connectors are worth adding first?

Start with the systems where your team asks the most questions. For most companies that is files and email, then chat, then tickets, then the CRM.

Google Workspace: Drive, Gmail and Calendar

Google's connectors sit at the top of the directory and are the easiest win if your company runs on Workspace. Per Anthropic's Google Workspace guide, Drive can search, read Docs, Sheets, Slides, PDFs and Office files, upload files and save Claude's output back to Drive. Gmail can search, draft, label and send. Calendar can find mutual availability and create or update events.

Good at: pulling a doc you half remember the name of, catching up on a long email thread, prepping for a meeting. Answers cite the emails, events or files they used.

Weak at: anything visual. Claude extracts text only from Drive files, and Gmail attachments are metadata only. Send, reply, forward, share and trash actions ask for approval by default. If you see "Access blocked: your institution's admin needs to review Claude for Google Drive," a Workspace admin has to mark Claude as a trusted app. Coworker's own Google Drive connector page covers the same data from a different angle.

Microsoft 365: SharePoint, OneDrive, Outlook and Teams

Anthropic's Microsoft 365 connector is available on every Claude plan and covers four services in one connection: document search across SharePoint and OneDrive, Outlook mail and calendar, and Teams chats and meeting information.

Good at: Microsoft shops that also want Claude. All access is delegated, so Claude reads or changes only what your account already can.

Weak at: setup. It requires a work or school account, a Microsoft Entra Global Administrator has to grant tenant consent, and write actions only work after administrators turn them on. Claude cannot attach files to the drafts it creates.

Slack

The Slack connector, built by Slack, searches public and private channels, threads, files and users, and can send messages and create canvases. Anthropic's Slack docs list the connector for Team and Enterprise plans and ask you to install the Claude app in Slack first.

Good at: "what did we decide about X" questions, since a lot of decisions only ever exist in a thread.

Weak at: noise. Slack search returns messages, not conclusions, so Claude often has to read many threads to answer one question, which fills the context window quickly. More on Slack and AI on Coworker's Slack connector page.

Atlassian: Jira and Confluence

Atlassian's Rovo MCP server is one connector for Jira, Confluence, Bitbucket, Loom and Atlassian's Teamwork Graph, per its directory listing. It can search with JQL, create, edit and transition issues, and create or update Confluence pages.

Good at: engineering and product teams. The governance is spelled out in the listing: Jira delete and project-admin tools are off unless an org admin enables them, and every tool call lands in your organization's audit log.

Weak at: anything outside Atlassian. Linking a Jira bug to the customer thread in Slack and the deal in Salesforce is still on Claude. See also Jira MCP options and Coworker's Jira connector.

Notion

Notion's connector can search, fetch, create, update and move pages, work with databases and read or add comments. It is strong when your wiki is well kept and weak when it is not, because Claude can only find what is written down and findable. Coworker has a separate post on the Notion MCP launch.

Salesforce and HubSpot

Salesforce's connector is listed as "Salesforce - Beta" in the directory, added August 2026, and is backed by Salesforce's Headless 360 MCP server. It answers pipeline and account questions while honoring Salesforce permissions. Being beta, expect changes. I compared the wider set of options in Salesforce MCP servers.

HubSpot's connector exposes 36 tools across contacts, companies, deals, tickets, campaigns and reports. It covers a lot, and that breadth has a cost covered below. More detail in HubSpot MCP.

GitHub, Linear, Box, Intercom and Zapier

• GitHub integration: adds repository files to a chat or project. It reads file names and contents only, with no commit history, pull requests or issues, per Anthropic's GitHub docs. For real coding work, Claude Code with MCP servers is the better fit.
• Linear: issues, projects, cycles and comments, built by Linear.
• Box: search plus Box's own AI Q&A and metadata extraction over your files.
• Intercom: search conversations and contacts; the tool list is read-focused.
• Zapier: runs actions across the apps in Zapier's catalog, useful when no native connector exists.

How do you add a connector to Claude?

The steps depend on your plan.

Individual plans (Free, Pro, Max). Go to Customize > Connectors, click the plus button, pick a connector and sign in to that service. Then turn it on per conversation from the plus menu under Connectors.

Team and Enterprise. Per Anthropic's connector guide, an Owner or Primary Owner first goes to Organization settings > Connectors, clicks Browse connectors and selects "Add to your team." That makes the connector available but grants nobody access: each member still authenticates with their own account. Two admin details matter:

• Member requests. People who cannot add connectors can request one. Requests are on by default on Team plans and off by default on Enterprise until an Owner turns them on.
• Enterprise-managed auth (beta). Authorize a connector once for the whole organization and members inherit access on first login.

Custom connectors. If a tool is not in the directory but has a remote MCP server, add it under Customize > Connectors > Add > Custom > Web with the server URL. Custom connectors work on Free, Pro, Max, Team and Enterprise, with Free limited to one. They are reached from Anthropic's cloud, not your laptop, so the server must be on the public internet, or reachable through Anthropic's MCP tunnels for private networks.

Are Claude connectors safe for company data?

Mostly yes, if an admin sets them up deliberately. The controls are real:

• Inherited permissions. Claude sees only what each person can see in the source system. The exception is a custom connector using a shared credential, which reaches whatever that credential can.
• Action restrictions. Owners can set each tool or tool category to Always allow, Needs approval or Blocked, org-wide. A common pattern is letting Claude read email but never send it.
• Domain lock (Enterprise). Owners can stop services on their verified domains from being connected to Claude accounts outside the organization.

And the caveats are real too. Connected services process data on their own infrastructure under their own terms, which may be outside the United States, even if your Claude Enterprise plan uses US-only inference. A Verified label is not a security audit. And any connector that reads outside content (email, shared docs, web pages) can carry prompt injection, so pair read access with approval on write tools. I go deeper on this in MCP security.

How many Claude connectors should you turn on?

Fewer than you think. Every enabled connector adds its tool definitions to the conversation, and HubSpot alone lists 36 tools. Anthropic's own guidance is to switch the Tool access setting from Auto to On demand "if you have 10 or more connectors active," to give conversations more room.

The cost is not just tokens. A context window crowded with tool schemas and raw search results leaves less room for the actual task, and answer quality drops as inputs grow. That is context rot. Enable the connectors a workflow needs, turn the rest off per conversation, and prefer connectors that return answers over ones that return piles of documents.

Where do Claude connectors fall short?

Each connector is a separate pipe. The Slack connector knows Slack, the Jira connector knows Jira, the Salesforce connector knows Salesforce. None of them knows that the escalation in #support, the P1 ticket and the renewal at risk are the same customer.

So a question like "which enterprise accounts reported this bug, and are any of them up for renewal this quarter?" turns into a chain of searches. Claude queries Slack, reads threads, queries Jira, reads tickets, queries the CRM, then tries to join it all inside one context window. That works for small questions. It gets slow, expensive and error-prone for the cross-tool questions teams care about most, and it starts from zero in every new conversation.

What is missing is a context layer: something that has already connected people, accounts, projects and decisions across tools, and hands Claude the answer instead of the raw material. Some vendors now ship that as a single connector. Glean, for example, lists a connector in Claude's directory that searches across the sources connected to Glean. I compared those approaches in Glean MCP and in my explainer on the context graph.

How does Coworker MCP fit with Claude connectors?

Coworker MCP is one connector that brings 50+ work tools into Claude as a single, permission-aware company graph. Instead of adding Slack, Jira, Drive, Gmail, HubSpot and Salesforce one by one, you connect Coworker once, and Claude queries OM2, Coworker's organizational memory.

OM2 is a context graph. It breaks messages, tickets, docs and meetings into facts linked to people, accounts and projects, and enforces source permissions when it answers, as described in the OM2 launch post. For Claude, that means a cross-tool question comes back as a resolved answer with sources rather than twenty search results to read.

The difference is measurable. Coworker's benchmarks ran 100 tasks across 7 categories on a Claude harness, comparing Claude with Coworker MCP against Claude with its native connectors to Google Drive, Gmail, Google Calendar, HubSpot, GitHub, Jira, Slack and Notion. On retrieval-heavy tasks, Coworker MCP used 9x fewer tokens (89.1% lower cost) and ran 64% faster; across all 7 categories it was 75.5% cheaper, and blind reviewers preferred its answers 84.5% of the time. On about 5% of questions the native-connector setup could not answer at all.

This is not an argument for removing native connectors. Keep them for actions inside one tool, like drafting a Gmail reply or editing a Notion page. Use Coworker MCP as the memory MCP layer for questions that span tools.

How do you set up Coworker MCP in Claude?

Per the Coworker MCP page, your admin enables MCP access for your company, and you add the Coworker MCP server in Claude and sign in with your existing Coworker login and SSO. Queries respect your existing permission model, the MCP tools are search and retrieval only, and setup typically takes under 30 minutes. The same server works in Claude Code, Cursor and other MCP clients, and the connectors page lists what Coworker connects to.

If you want to see it on your own stack, book a demo, or ask for the benchmark to be run on your company's data.

Frequently asked questions

What connectors does Claude have?

Claude's Connectors Directory listed 974 connectors on October 19, 2026. The most used include Google Drive, Gmail, Google Calendar, Microsoft 365, Slack, Notion, Canva, Figma, HubSpot, Asana, Linear and monday.com, plus Atlassian's Jira and Confluence server and a Salesforce beta. You can also add any remote MCP server as a custom connector.

Does Claude have connectors like ChatGPT?

Yes. ChatGPT also connects to outside apps, including remote MCP servers, though its setup screens and admin controls differ from Claude's. My ChatGPT MCP guide covers the ChatGPT side.

Is it safe to use connectors in Claude?

It is as safe as the controls you set. Claude inherits each user's permissions, and Team and Enterprise Owners can block or require approval for write tools org-wide. Verified labels are not security audits, third-party services handle data under their own terms, and connectors that read outside content carry prompt-injection risk.

Can I use custom connectors on the free Claude plan?

Yes, with a limit. Custom connectors using remote MCP are available on Free, Pro, Max, Team and Enterprise, and Free users are limited to one custom connector.

Who can add connectors on Claude Team and Enterprise?

An Owner or Primary Owner adds connectors for the organization in Organization settings > Connectors. Each member then authenticates individually. On Enterprise, a member whose custom role includes managing the organization's libraries can add one too.

What is the difference between Claude connectors and MCP?

MCP is the open protocol. A connector is a specific MCP server packaged for Claude, usually listed in the directory with a review label. Any directory connector or custom connector runs on the same infrastructure. For how the protocol itself works with Claude, see my Claude MCP guide.

Do Claude connectors work in Claude Code?

Yes. Anthropic's docs say the same directory catalog serves Claude Code, and connectors also work through the API via the MCP connector.

Can Claude write to my connected tools?

Yes, when the connector exposes write tools and your permissions allow it. Gmail, Calendar and Drive actions ask for approval by default, Microsoft 365 writes need admin enablement, and Owners can block write tools for everyone.

Related reading

• Claude MCP: How to Connect Claude to Your Tools
• What Is MCP (Model Context Protocol)?
• MCP Security
• Glean MCP
• What Is a Context Graph?
• What Is a Context Layer?
• What Is Context Rot?
• Salesforce MCP
• Coworker benchmarks