Last updated: November 9, 2026
TL;DR
The best Claude connectors for a small team are the ones that match the tools it already uses every day: the email, calendar and files suite first, then the CRM and the project tracker, then Canva, Figma or GitHub for the people who design or code. Before anyone connects, an Owner should block every tool that sends, deletes or pays, and put the other write tools behind Needs approval. Custom connectors go last, because Anthropic hasn't reviewed them.
- 📬 What to turn on. One connector per job: Google Workspace or Microsoft 365, one CRM, one tracker, and design or code tools only for the people who need them. Anthropic's directory listed 993 connectors on October 21, 2026, so the hard part is choosing.
- 🔍 What Claude can see. Only what each person can already open in the connected service, and only in conversations where the connector is switched on.
- ✋ What Claude can do. An Owner sets each tool to Always allow, Needs approval or Blocked, for the whole organization. Reads can run freely. Block sending, deleting and paying, because in Cowork a Needs approval tool pauses only in Manual mode.
- 🧰 What to add last. Custom connectors and desktop extensions. Anthropic hasn't reviewed custom connectors, only Owners can add one on Team, and desktop extensions are unrestricted until you switch on the allowlist.
The best Claude connectors to turn on first
Turn on one connector for each job your team does every day, and stop there until people actually use them. For most small teams that means five or six: an email, calendar and files suite, a chat tool, a CRM, a project tracker, and design or code tools for the few people who need them.
On the Claude Team plan, an Owner adds each connector for the organization, and every member then signs in with their own account. Adding one "doesn't automatically grant anyone access," per Anthropic's connectors guide. To test a connector first, the Owner can choose Connect just for me. Our own Team console has 25 connectors enabled, including Apollo.io, Asana, Atlassian, Attio, Box, Calendly, Canva, Figma, GitHub Integration and Gmail.
The tables come from each listing in Anthropic's directory and Anthropic's docs, read on October 21, 2026. Set permissions for everything in the "can change" column.
Email, calendar and files
Pick the suite your company already runs on, Google Workspace or Microsoft 365, not both.
| Connector | Made by | Claude can read | Claude can change |
|---|---|---|---|
| Gmail | Search and read threads | Drafts, replies, sends, forwards, labels, filters, trash | |
| Google Calendar | Events, attendees, free time | Creates, updates and deletes events, responds to invites | |
| Google Drive | Docs, Sheets, Slides, PDFs, Office files, who a file is shared with | Creates, updates, copies, shares and trashes files | |
| Microsoft 365 | Anthropic | SharePoint, OneDrive, Outlook mail and calendar, Teams chats | Nothing until an admin turns on write actions, then mail, events, SharePoint files and Teams messages |
| Slack | Slack | Public and private channels, threads, users, canvases | Sends messages, creates and updates canvases |
| Box | Box | Search, file content, AI questions across files | Uploads files and versions, creates folders |
The Google descriptions conflict. Anthropic's Gmail page says Claude "can't create, send, or modify messages," and its Calendar page calls that connector read-only. The directory listings, made by Google and added in March 2026, show 29 Gmail tools, including send_message, forward and trash_thread, and Calendar tools including delete_event. Go by the tool list on the connector's page in your own console.
Microsoft 365 is the stricter of the two. A Microsoft Entra Global Administrator must grant tenant consent, and write actions stay off until your admins switch them on, per Anthropic's Microsoft 365 page.
CRM and sales
Connect the CRM your team already keeps current.
| Connector | Claude can read | Claude can change |
|---|---|---|
| HubSpot | Contacts, companies, deals, tickets, campaigns, email and content analytics | CRM records, pipelines, marketing emails, landing pages, blog posts (36 tools) |
| Attio | Records, notes, emails, call recordings and transcripts | Records, notes and tasks |
| Apollo.io | People and company search, enrichment, hiring and funding signals | Contacts, deals, tasks and email sequences, which it adds contacts to only with your confirmation |
| Calendly | Event types, availability, bookings | Creates event types and bookings, cancels events |
Salesforce reached Claude on September 15, 2026 as Salesforce in Claude (beta), a plugin rather than a plain connector, per Anthropic's release notes. Finance connectors such as QuickBooks and PayPal ship with Anthropic's small-business plugin, which our Claude for Small Business breakdown covers.
Project management
Connect the one tracker your team works from. All five can write as well as read.
| Connector | Claude can read | Claude can change |
|---|---|---|
| Asana | Tasks, projects, portfolios, goals | Creates projects and tasks, updates tasks |
| Atlassian | Jira, Confluence, Loom, Bitbucket | Creates and edits Jira issues and Confluence pages |
| Linear | Issues, projects, cycles, documents | Creates and updates issues, projects and comments |
| monday.com | Boards, items, dashboards | Creates and updates items, posts updates, builds automations |
| Notion | Pages and databases | Creates, updates, moves and duplicates pages and databases, comments |
Atlassian adds its own guardrails. Jira delete and project-administration tools "are off unless your organization admin enables them," and every tool call lands in Atlassian's audit log. Claude's audit logs are Enterprise-only, so on Team that is the record of what Claude did in Jira.
Design
Turn on Canva for whoever makes marketing graphics and Figma for whoever owns product design. Nobody else needs them.
| Connector | Claude can read | Claude can change |
|---|---|---|
| Canva | Designs, pages, content, folders | Generates, imports and exports designs, moves items, comments |
| Figma | Design context, variables, components, metadata | Generates diagrams in FigJam |
In our console, Canva and Figma carry the Interactive badge, as do Asana and Box. An interactive connector renders a live view, such as a task board, inside the conversation.
Development
The built-in GitHub Integration reads code and nothing else: file names, contents and branches, but not commits, pull requests or issues, per Anthropic's GitHub page. Developers who want Claude to edit code and open pull requests use Claude Code, which picks up claude.ai connectors when they sign in with their Claude account. Owners can allow or deny MCP servers for every developer in Claude Code's managed settings. Our compound engineering guide shows how we keep that setup consistent.
What data can Claude access through connectors?
Claude can access only the data the signed-in person can already open in the connected service, and only in a conversation where that connector is switched on. "Claude inherits each person's permissions from the connected service," per Anthropic's connectors guide. A setting in Claude can narrow that access but never widen it.
So Claude never sees all data in the connected service. Five limits to know before rollout:
- 🔑 Shared logins reach more. A custom connector that uses one shared credential reaches whatever that credential can access, for every member who uses it.
- 📁 Shared Projects can't pull files. You can add Google Drive files and GitHub repositories only to a private project. In a shared project the option is dimmed, per the Drive page.
- 👀 Some content stays invisible. Claude can't read comments or suggested edits in Drive files, or images embedded in email.
- 🌍 The service keeps its own rules. "Connected services process data on their own infrastructure, under their own terms," and settings such as US-only inference don't change where they run.
- 🔬 Research skips approvals. During Research, Claude can call connector tools without asking again, so Anthropic's custom connector guide says to disable write-capable tools first.
How to set each tool to Always allow, Needs approval or Blocked
An Owner sets each connector tool, or each group of tools, to Always allow, Needs approval or Blocked, and the setting applies to everyone. "Individual users can't override it," per Anthropic's connectors guide. Allow reads, put record changes behind Needs approval, and block anything that must never run unattended.
Do it per connector, before anyone connects:
- Go to Organization settings, then Connectors, and open the connector.
- Set the read-only group to Always allow.
- Set the write group to Needs approval.
- Go through the write tools one by one and mark anything that sends, deletes, refunds or pays as Blocked, unless someone needs it this week.
- In Organization settings, then Cowork, leave Allow "Always allow" for connector tools off. It is off by default, per Anthropic's Cowork admin article. While it's off, members can't skip per-task approval for write tools in Cowork.
- On the same page, decide on Allow "Automatically approve" mode. It is on by default. Turning it off removes Auto from members' mode selector.
Step 6 matters because Needs approval doesn't always wait for a person. In Cowork, a Needs approval tool pauses the task only in Manual mode. In Auto mode Claude decides on write and delete tools itself, and in Skip mode they are approved, per Anthropic's Cowork getting started article. Only Blocked holds in every mode, which is why sending, deleting and paying belong there.
These are the tools we'd set first, using the names from each directory listing:
| Tool | Connector | Setting |
|---|---|---|
search_threads, get_thread | Gmail | Always allow |
send_message, forward | Gmail | Blocked |
trash_thread | Gmail | Blocked |
delete_event | Google Calendar | Blocked |
share_file | Google Drive | Needs approval |
trash_file | Google Drive | Blocked |
slack_send_message | Slack | Blocked, or Needs approval if Cowork stays on Manual |
manage_crm_objects | HubSpot | Needs approval |
create_refund, pay_order | PayPal | Blocked |
executeDestructive | Atlassian | Blocked |
One detail catches people. Cowork skips approval for read-only tools only when the connector marks them read-only, and most custom connectors don't, so all their tools wait for approval.
Custom connectors and the desktop extension allowlist
Add custom connectors and desktop extensions last, and only from vendors you trust. Anthropic hasn't reviewed a custom connector, which is any remote MCP server added by its URL. On Team, only Owners and Primary Owners can add one. Desktop extensions run locally in Claude Desktop, and members can install any of them until you switch on the allowlist.
Every connector carries one of three labels, defined on Anthropic's verification page:
| Label | What Anthropic did |
|---|---|
| Verified | Tested its tools for quality and compatibility. "It isn't a security audit" |
| Community | Screened it before listing, but didn't review it in depth |
| Custom | Nothing. You added it yourself |
The label changes how a connector is shown, not what it can do, and the developer can change its tools after review.
To add a custom connector for your organization:
- Go to Organization settings, then Connectors, select Add, then Custom, then Web.
- Enter the remote MCP server URL.
- Choose how people sign in: Sign in now, Sign in when needed or No sign-in.
- Select Add. Members then find it under Customize, then Connectors, with a Custom label, and connect their own account.
Three limits come from Anthropic's custom connector docs. The server must be reachable over the public internet, so a server inside your network needs Anthropic's IP ranges allowlisted or an MCP tunnel. Authentication settings can't be changed after you add it; you remove it, add it again, and everyone reconnects. And a custom connector can turn on you: "Malicious MCP servers may include hidden instructions that try to make Claude perform unintended actions," per Anthropic's custom connector guide.
The desktop extension allowlist is off by default, and it was off in our console. Until you enable it, "users will be able to access all desktop extensions in the registry," per Anthropic's allowlist article. Switching it on force-deletes existing extension installs from members' Claude Desktop and blocks anything not on the list. Add the extensions people need first (Organization settings, then Connectors, the Desktop tab, Browse extensions, Add to your team), warn the team, then switch it on.
Member requests are the last setting. "Let members request connectors" is on by default on Team, and only Owners and Primary Owners can change it. Requests land under Organization settings, then Notifications, then Requests, where you can approve, decline, or decline and block the connector.
Frequently asked questions
What are the best Claude connectors?
The best Claude connectors are the ones for tools your team already uses daily: Google Workspace or Microsoft 365 for email, calendar and files, then Slack, your CRM (HubSpot, Attio or Apollo.io) and your tracker (Asana, Atlassian, Linear, monday.com or Notion). Add Canva, Figma or GitHub only for the people who design or code.
What data can Claude access through connectors?
Claude can access only what the signed-in person can already open in the connected service, and only in conversations where the connector is switched on. It inherits each person's permissions. Settings in Claude can narrow that access but never widen it.
Can Claude access all data in the connected service?
No. Claude sees only the data your own account in that service can see. The exception is a custom connector that uses one shared credential, which reaches whatever that credential can access, for everyone who uses it.
What are Claude custom connectors?
A custom connector links Claude to a remote MCP server by its URL, for a tool that isn't in Anthropic's directory or that your team built. Anthropic hasn't reviewed custom connectors. On the Team plan, only Owners and Primary Owners can add one, and each member then connects with their own account.
Who can add connectors on a Claude Team plan?
Owners and Primary Owners add connectors for the organization under Organization settings, then Connectors. Members then sign in with their own accounts. Members who can't add connectors can request one, which is on by default on Team.
Can Claude send emails through the Gmail connector?
Anthropic's docs disagree. Its Gmail help page says the connector is read-only, but the directory listing made by Google shows tools such as send_message, forward and trash_thread. Check the tool list on the connector's page in your console, and block sending unless someone needs it.
Are Claude connectors safe?
Verified connectors have had their tools tested by Anthropic, but verification isn't a security audit. Community connectors are screened, not reviewed in depth, and custom connectors aren't reviewed at all. Connect only to developers you trust, and block tools that send, delete or pay.
What is the difference between a connector and a desktop extension?
A connector usually reaches a remote service over the internet and works on web, desktop, mobile and in Claude Code. A desktop extension is a local MCP server packaged as an MCP Bundle that runs on your computer inside Claude Desktop, with your user permissions.
How many connectors does Claude have?
Anthropic's connectors directory listed 993 connectors on October 21, 2026, from Anthropic and outside developers. Any remote MCP server can also be added as a custom connector by its URL.
Get Claude Team set up properly
We set up Claude Team for small businesses: the seat plan, the security settings, shared Projects and skills written from your actual processes, and the first scheduled routines. If you'd rather have it done once and done right, see how our Claude Team setup works or book a 30-minute call.
Stop guessing. Score your first workflow.
Score one workflow in about five minutes. Nine questions, a grade, and the specific gap holding it back - no call required.
Related Posts
Cursor worktrees vs Claude Code worktrees: four differences that matter
Cursor worktrees vs Claude Code worktrees: both give each agent its own git checkout. How each starts one, sets up the checkout, brings changes back and cleans up old worktrees, the pattern each covers that the other doesn't, and which fits how you work.
Claude Code worktrees: run parallel sessions without collisions
How to use Claude Code worktrees: start a session in its own git worktree with claude -w, copy .env files in with .worktreeinclude, branch from your current work or a pull request, isolate subagents, clean up, and fix the errors worktrees print.
Claude Code put account emails in User-Agent headers: what to update and check
Claude Code sometimes put signed-in users' account email into the User-Agent header of requests it wrote. Version 2.1.317 adds a check that asks first, the stable channel was still on 2.1.311, and the next release removes the email from the model's context. What to update, and what the check misses.