Claude Cowork Connectors: What Actually Connects in 2026
Cowork does not have its own connector list. That is the first thing to get straight, because people go looking for one. Anthropic’s connector documentation says web connectors are available “for all users on Claude, Cowork, Claude Desktop, and Claude Mobile,” and the directory’s own filter offers exactly two targets, Claude and Claude Code. Cowork is a surface on Claude, so whatever you connect in Claude is what a Cowork task can reach.
What changes in Cowork is not what connects. It is that the agent uses those connectors unattended, across many steps, which makes the permission layer matter far more than it does in chat.
The built-in connectors
These are first-party and take one click, no MCP server to run:
| Connector | What it reaches | Write capability |
|---|---|---|
| Gmail | Search, read, labels, threads, drafts | Draft, send, reply, forward, with approval |
| Google Calendar | Events, shared calendars, availability | Create, update, delete, RSVP, recurring |
| Google Drive | Docs search, metadata, permissions, recent changes | Share, move, trash, upload, create folders |
| Microsoft 365 | Outlook mail, calendar, OneDrive, SharePoint | Write tools exist but ship off by default |
| GitHub | Repositories and code context | Repository actions |
| Slack | Channels and messages | Included on Team plan seats |
| 1Password | Vault items | Varies by item permission |
Two of those rows deserve a note. The Gmail connector’s capability badge in Anthropic’s directory reads Read & write, and the page states that Claude “can also draft, send, reply to, and forward emails, and asks for your approval by default before anything goes out.” Claude’s Gmail side sends now. That is a recent change and a lot of older writing on the web, ours included, still describes it as draft only.
The Drive row is the opposite kind of surprise. Drive can read Sheets, Slides, PDFs, images and Office files, and it can upload and share, but editing an existing Google Sheet cell by cell is not one of its tools. Cowork’s answer to spreadsheet work is to build an .xlsx with live formulas and save it to Drive.
The directory, and how to read a listing
Beyond the built-ins sits Anthropic’s connectors directory, powered by MCP and running to several hundred entries across categories like productivity, communication, data, financial services and life sciences. Airtable, Amplitude, Asana, Apollo.io, Ahrefs, Affinity, Adobe Workfront and Twilio are all in there.
Each listing carries a capability badge, and the badge is the only part worth reading closely:
- Read means context only. Twilio’s listing is a clean example: 1,800+ endpoints of API documentation search. It will tell Cowork the endpoint for sending a WhatsApp message. It will not send one.
- Read & write means the connector exposes actions.
- Interactive means it can render a live interface inside the conversation.
A connector being listed tells you the app can be reached. It does not tell you the agent can change anything in it.
Custom connectors, and the free-tier detail people miss
Anything outside the directory connects over remote MCP. Anthropic’s documentation is explicit about who can do this: custom connectors “are available on Claude, Cowork, and Claude Desktop for users on free, Pro, Max, Team, and Enterprise plans,” with free accounts limited to one.
You add one under Customize → Connectors → Add custom connector with a name and URL, plus OAuth client ID and secret if the server needs them. One constraint catches self-hosters: custom connectors are called from Anthropic’s cloud, not from your machine, so the server has to be reachable on the public internet.
The permission layer, which is where Cowork differs from chat
Per connector, under Customize → Connectors → Tool permissions, every tool sits in one of three states: Always allow, Needs approval, or Blocked. Owners on Team and Enterprise can set those org-wide, and individual members cannot override them.
Cowork then layers its own mode on top, chosen from the mode selector in the message box:
- Manually approve. Claude pauses on each action.
- Automatically approve. Claude keeps going, screening each action for prompt injection and exfiltration and blocking what it judges unsafe. Anthropic notes this mode consumes more of your usage limit, because the checking costs tokens.
- Skip all approvals. Nothing is checked.
Auto mode is currently Pro and Max only. On Team and Enterprise, an organization can require per-task approval for write-capable tools, which overrides an individual’s “Always allow” preference.
One practical tuning note: with ten or more connectors active, switch Tool access from Auto to On demand. Loading every tool schema into a long agentic run eats context you would rather spend on the work.
Where the whole model stops
Connectors feed a run that has already started. None of them start one.
There is no such thing as an MCP trigger. The protocol's own working group states plainly that clients find out about server-side changes by polling for them, and a real event mechanism remains unshipped.
Cowork’s scheduled tasks close part of that gap, and genuinely so: type /schedule in any task and it runs daily, weekly or monthly in Anthropic’s cloud with no device online. But a clock is not an event. Nothing in the connector model says “when this invoice arrives, go.”
Pointing Cowork at something that does have triggers
Carly runs an MCP server at https://carlyassistant.com/mcp/, which you add in Claude under Customize → Connectors as a custom connector. That gives a Cowork task direct use of Carly’s email, calendar, contacts, CRM, workflow, file, notes and booking tools, so a run can finish an action rather than stop at an approval prompt.
The reason to bother is what sits on the other side. Carly’s own automation starts on events: an inbound email, a booking, a form submission, a changed CRM record. It holds several mailboxes at once, each authorized separately, so a work account and a personal one both stay connected instead of replacing each other. It reaches your stack through 260+ native integrations plus your own API key for anything else with an API, pasted on the integrations page. Zapier-style workflow steps are free, and AI agents start at $35 a month.
So the honest split: connect Claude’s directory for context, and put the thing that has to fire on its own behind Carly.
FAQ
Does Claude Cowork have its own connectors?
No. Cowork uses the same connectors as Claude. Anything you connect in Claude’s settings is available to a Cowork task, and the directory filters by Claude or Claude Code rather than by Cowork.
Can Claude Cowork send email?
Yes on Gmail, where the connector drafts, sends, replies and forwards with your approval by default. On Outlook it can send through the Microsoft 365 connector’s write tools, which are off until an administrator enables them.
Can Claude Cowork edit a Google Sheet?
Not cell by cell through the Drive connector, which reads Sheets rather than editing them. Cowork builds Excel files with working formulas instead and can save them to your Drive.
Do custom connectors need a paid plan?
No. Custom remote MCP connectors work on Free as well as the paid tiers, with free accounts limited to one. The paid plan is what unlocks Cowork itself.
How many connectors can I have active at once?
There is no hard cap, but Anthropic recommends switching tool access from Auto to On demand once you pass ten active connectors, so the tool schemas stop crowding your context.
More: Claude Cowork pricing · Claude Cowork plugins · Claude Cowork skills · Claude Cowork limitations · ChatGPT Work connectors
Ready to automate your busywork?
Carly schedules, researches, and briefs you—so you can focus on what matters.
See what people say
"Before Carly, I relied on a Calendly link, but the whole process felt impersonal and not very professional. Carly changed that by handling all the back-and-forth, so I'm no longer stuck in endless email threads trying to line up schedules.
Now Carly reaches out to candidates, shares my real-time availability, lets them pick a slot, then sends a Zoom link and drops it straight into my calendar. She sends reminders to both of us before each call, which has significantly reduced no-shows and last-minute confusion.
On top of scheduling, Carly acts like a full executive assistant, sending me my schedule the night before so I can prepare for each call. It reminds me of the old x.ai assistant, but Carly is noticeably smarter, faster, and better suited to my healthcare recruitment business."
