Connectors let Claude read from and act in the apps you already use: your inbox, calendar, docs, task tracker and design tools. That's where Claude stops being a chat window and starts doing real work. It's also where it can send, edit or delete things on your behalf. Here's what to connect first, and the settings to change before you rely on it.
Three things to know first
- It uses your permissions. Claude can only reach what your account can reach in that app, and it can take actions your account is allowed to take.
- It works across surfaces. The same connectors work in Claude on web, desktop and mobile, in Claude Code, and through the API.
- Claude picks them automatically. Once an app is connected, Claude can use it when it fits your request, without you naming it.
Connect one
- Go to Customize → Connectors and select +. You can also type / in a chat and open Connectors.
- Pick a connector and review its description, and whether it can read, write, or both.
- Select Connect and sign in to the app.
- In a chat, use + → Connectors to toggle which ones that conversation can use.
On Team and Enterprise, an owner has to enable a connector for the organization first. Then each person still signs in with their own account.
Pick by job, not by hype
Start with the one job you repeat most. Add more only when a real task needs them.
| Job | Connect | Try |
|---|---|---|
| Plan your day | Calendar and email | "What needs my attention today? Draft replies but don't send them" |
| Find your own knowledge | Google Drive or Notion | "Find what we decided about pricing and cite the doc" |
| Run projects | Linear, Slack or GitHub | "Turn this bug report into an issue with steps to reproduce" |
| Make visuals | Canva | "Make three social graphic options from this headline" |
| Research audio | A podcast transcript connector | "Find where this topic comes up in recent episodes" |
My own non-obvious favorite is Audioscrape, which searches already-transcribed podcasts and pulls speaker-labeled transcripts. More on that in my AI information filter.
Set permissions before first use
Under Customize → Connectors, open a connector to see Tool permissions. Each tool is grouped as read-only or write/delete, and you can set it to:
| Setting | Use for |
|---|---|
| Always allow | Reading and searching |
| Needs approval | Anything that sends, creates, edits or deletes |
| Blocked | Actions you never want, like deleting emails |
A good default: always allow reads, require approval for writes. Content Claude reads through a connector, like an email or a shared doc, can include instructions written by someone else, and approval is your checkpoint before anything leaves your account.
Team and Enterprise owners can set these restrictions org-wide, and they can only narrow what your account can already do.
Keep conversations lean
Every connected tool takes up room in the conversation.
- Toggle per chat. Turn on only what a conversation needs under + → Connectors.
- Use Tool access. Under + → Connectors → Tool access, Auto works for most people. With 10 or more connectors, switch to On demand.
- Disconnect what you don't use. It's less clutter and less access.
For the trade-offs between connectors and command-line tools, see MCP vs CLI.
Custom connectors
Any service with a remote MCP server can be added: Customize → Connectors → + → Add custom connector, then enter its name and URL.
- Free plans get one custom connector.
- It must be public. Claude connects from Anthropic's cloud, not your laptop, so a server on a private network won't work without extra setup.
- They aren't verified by Anthropic. Only add servers from organizations you trust.
Start here
- Connect one app for your most repeated job
- Set writes to Needs approval before first use
- Try one task from the table above
- Turn off connectors a chat doesn't need
- Review and disconnect unused connectors monthly
Go deeper: MCP vs CLI · Claude, End to End · Stop Saving Everything: An AI Filter