Claude Duet connectors are the ones you've already connected in your Claude account: Duet uses them, along with your Claude memory and skills, so you add and manage connectors in Claude's settings, not in Duet. Anthropic's custom connector guide doesn't list Duet yet, so whether a custom MCP connector you add in Claude also works in Duet isn't documented. Duet is a research preview for Pro and Max in the US, by waitlist, and we're on that waitlist: unlike the other personal agents we write about, we haven't run this one yet, so everything below is from Anthropic's pages, with what we'll test first marked at the end. Docs checked 2026-10-21.
What is Claude Duet, and who can use it?
Anthropic's proactive agent app: you hand it work through messages, it briefs you each morning, and it checks your connected apps through the day. It opened as a research preview on October 21, 2026, by waitlist, "starting with US Pro and Max subscribers" (Anthropic).
- Where it runs: a standalone app on iOS and Android, and on claude.ai, "separate from the main Claude app", with its own virtual machine and browser, so your laptop and phone can be off.
- Model: Claude Opus 5.5, no model choice (Claude help).
- Plans and cost: Pro and Max in the US, invitations in waves, included in your plan during the preview but within its limits. Background work such as the morning brief and check-ins counts, and you get one full limit reset on admission, available until November 30, 2026. Not on Team or Enterprise yet.
- Approvals: it waits for you before it sends a message, pays, signs or deletes anything.
Does Claude Duet use your Claude connectors?
Yes. Duet "works with the connectors you've connected in Claude," such as Gmail, Google Calendar, Google Drive and Slack (Claude help), and you manage them in Claude's settings. Slack works with limits, Microsoft 365 isn't available yet, plugins such as the small business plugin work, and Link and Shop Pay are connected inside Duet. Custom connectors are the open question: Anthropic's custom connector guide lists Claude, Cowork and Claude Desktop, not Duet. Disconnecting a connector stops new access only; Duet keeps what it already learned.
What does Claude Duet remember?
Your Claude memory. Duet "also uses and adds to your memory" (Claude help), separate Duet channels share that same memory, and you see or edit it under Settings > Memory in Claude, where every topic is listed. Duet also searches your Duet chats and your Claude chats outside of projects, which you can stop with Search and reference chats off. Each Claude project has its own memory, and Duet "can't see all of your projects or the files on your computer" (Claude help).
That's a different design from the other personal agents. Grok Bot keeps memory per Bot, a dot keeps notes you can't open, and Muse keeps files on its own machine. Duet's memory is simply Claude's, editable in one place. It also means Duet inherits whatever is already in your Claude memory, good and stale alike.
What can't Claude Duet do with your context?
Duet can read your Claude work but can't change it, and its memory, chats and finished work stay in your account, where no teammate or other vendor's agent can read them. One early reviewer, Claire Vo, named the gap: "missing desktop and multiplayer support keeps it out of my working stack" (Lenny's Newsletter).
| What you need | What Duet does natively | Gap |
|---|---|---|
| Your Claude memory, skills and connectors | Yes, from the first message | One Claude account only |
| Background work | Morning brief, checks at set intervals | Urgent items can land between checks; checks use your limits |
| Update your Claude projects | Reads your Claude work | Can't change it |
| Teammates see its work | No Team or Enterprise access yet | Single user |
| dots or Grok Bot read its work | No documented bridge | A shared store |
| Custom MCP connectors | Uses connectors set up in Claude | Not documented for Duet |
Claude Duet vs dots vs Muse: where does the context live?
In three different places. Duet runs on your Claude account and its memory is Claude's. A dot gets its memories from ChatGPT and keeps its own notes you can't view or edit (OpenAI help). Muse keeps its memory as files on "an isolated linux box with a browser" that Meta says you can "inspect, edit and download" (Meta). All three ask before acting (in Muse, a component called Sentinel is "the sole permission authority for connector actions"), and none documents a way to read another's memory. Meta says "Muse also supports custom connectors so you can plug in services we don't support yet" (Meta); Duet uses what's in your Claude account. The four agents side by side, with our picks, are in personal AI agents compared.
What we'll test first when our invitation arrives
Two things, and we'll update this post the same day. First, whether the Tokenrip connector we already have in Claude shows up in Duet. Duet uses "the connectors you've connected in Claude", and Tokenrip is one of ours, so if custom connectors are included, Duet reads the same workspace our Claude Codes, Grok Bots, dot and ChatGPT already do, with no setup at all. Second, if it doesn't, whether Duet's own virtual machine can take the setup line below the way our Grok Bots and dot did: both have their own computer, both installed the Tokenrip CLI there and signed in, and that's the route we prefer for any agent with a shell. Then the interesting test: whether Duet's morning brief and scheduled check-ins can read the workspace, since a proactive agent that sees the open tasks and last night's handoffs is a much better morning brief than one that only sees your inbox.
Until then, Tokenrip and Duet meet through Claude itself: Tokenrip in a Claude chat reads and adds to the same Claude memory Duet uses. To connect Claude, add a custom connector with https://api.tokenrip.com/mcp and the emailed code; to let an agent with a shell do the setup, give it this line, and you give an email address and a six-digit code.
Connect me to Tokenrip. Fetch https://tokenrip.com/setup?ref=claude-duet and follow it.