Connectors in chat
Use your connected apps in any chat: switch them on from the + menu, point the AI agent at one app or account with @, and see every action it takes in the chat.
Updated 2026-10-20
Every connector you switch on works in every chat. Each connector has one switch: the switch in a chat's + menu is the same switch as on its card in Settings → Connectors, so turning a connector off in one chat turns it off everywhere.
A connector lets the agent use an outside app or service, such as Gmail, Notion or your browser. See Connectors for the full picture.
Before you start
- At least one connector added. See Connect an app, Built-in connectors or Custom connectors.
- An AI model connected to the chat. See Connect an AI model.
Switch a connector on or off from a chat
- Click the + button next to the message box.
- Click Connectors. The number beside it is how many connectors this chat can use right now.
- Flip the switch on a connector's row.
The change applies to every chat, and to Settings → Connectors, straight away.
The Connectors panel
- Search: click the magnifier next to CONNECTORS and type to filter the list. Press Escape to close the search.
- The note line at the top reads: "Switched-on apps work in every chat. Turning one off turns it off everywhere. Click a name to mention it."
- Sections, in this order:
- Custom: servers you added yourself.
- Granular: Granular's built-in connectors and your connected app accounts.
- This computer, with Claude only, Codex only and Gemini only: connectors that come from your own Claude, Codex or Gemini setup. The section for the chat's own model comes first. See This computer connectors.
- Each row shows the app's logo and name, a small mark for where it comes from, and its switch. Hover over the switch to see where the connector comes from and whether it's on.
- The amber "!" means the connector's sign-in never finished, so it's left out of chats. Hover to read "Needs sign-in", and click it to open that connector's settings and finish signing in.
- Hidden rows: connectors that can't run on this chat's model aren't listed. A faint line counts them, for example "2 more work only in Claude chats". Hover over it to see their names.
- Manage connectors → at the bottom opens Settings → Connectors, where you add, remove and set up connectors.
Claude in Chrome is listed under This computer. To let the agent act in your browser, turn it on with its switch.
Limit a connector to some projects
Every connector works in All projects at first. To keep one to certain projects, open it in Settings → Connectors and change Available in. Chats in other projects don't get it, even when it's switched on, and it isn't listed in their Connectors panel. See Connector accounts.
Point the agent at one app with @
When you have more than one copy of an app, for example two Gmail accounts, you can tell the agent exactly which one to use.
- In the message box, type @. A list opens with three tabs: All, Connectors and Skills. Each tab shows how many items match.
- Type a few letters to narrow it. Press Tab (or Shift+Tab) to move between the tabs, or click one. The Connectors tab lists only your connectors.
- Pick the one you mean with the arrow keys and Enter, or click it.
- Finish your message and send it.
The Skills tab lists skills installed for this folder; picking one adds it to your message the same way. See Resource library and skills.
You can also click a connector's name in the Connectors panel to add it to your message.
The mention appears as a chip in your message, tied to that exact account. When you send, Granular tells the agent to use that connector for this request.
A mention never switches anything on. If the connector is switched off, or can't run on this chat's model, the mention is just text and the agent isn't told to use it.
If you type a slash and an app's name, for example /gmail, and no command matches, Granular reminds you that Gmail is a connector and to use @Gmail instead.
What you'll see during a turn
- Connecting your tools (N): in Claude chats, this shows for a few seconds while your connectors start, before the agent answers.
- A card for each connector call: the connector's icon (the same one its card shows in Settings → Connectors), its name and account, for example "Gmail · Work", and a pulsing dot while it runs. It then shows a ✓ when the call worked or a ✗ when it failed, with the action's name. If a connector has no icon of its own, the card shows a plain tile with its first letter.
- Connected apps name the app and the action in plain words, for example Send email. A call that runs several actions shows one line per action, each with its own tick or cross.
- Tool calls · N: before using an app, the agent often runs a few quick steps to find the right action. These fold into one faint Tool calls row with a count. Click it to see each step.
- Bridge cards: in a Codex or Gemini chat, a bridged call shows as Bridge to Claude · app · account, for example "Bridge to Claude · Gmail · Work". Click it to see Claude's own steps. See Bridge.
- "Tool": if Granular can't tell which connector a call belongs to, the card just says Tool rather than guess.
Granular's own actions, such as running a command in the terminal or opening a preview, have their own rows in the chat. They never show as a connector card.
- Approval cards: if an app account is set to ask you first, the agent stops and a Permission · app action card asks before it sends, changes or deletes anything. Click Allow or Decline. See App permissions.
If a connector isn't signed in
A connector whose sign-in never finished is left out of your chats, so it can't stall the agent while it waits for a sign-in. After your first message in a session, a card names it:
One tool is not signed in (or 2 tools are not signed in)
"These are left out of this session so they can’t stall it:" followed by their names, then "Connect them to use them here, or ignore this and keep working without them."
- Connect opens the connector so you can finish signing in.
- Dismiss hides the card.
- Tick Don't show this again. Keep skipping these tools until I connect them. to stop the card appearing for these connectors. The button then reads Confirm. The card comes back if one of them is signed in and later signed out again.
The card shows at most once per session. When you finish signing in, it goes away.
Which AI model can use what
| Connector | Claude | Codex | Gemini |
|---|---|---|---|
| Built-in connectors | Yes | Yes | Yes |
| Custom connectors | Yes | Yes | Yes |
| Connected apps | Yes | Yes | Yes |
| Claude account connectors | Yes, when you're signed in with your Claude account | Through Bridge | Through Bridge |
| Claude in Chrome | Yes, when you're signed in with your Claude account | No | No |
| Claude Code servers | Yes | Through Bridge | Through Bridge |
| Codex apps and Codex settings | No | Yes | No |
| Gemini CLI servers | No | No | Yes |
Claude account connectors and Claude in Chrome need the Claude model signed in with your Claude account. A Claude chat that runs on an API key, Granular Hosted or a custom provider doesn't get them, and they aren't listed there.
Bridge lets a Codex or Gemini chat use a Claude-only connector by asking your own Claude to do the task. See Bridge.
When changes take effect
- Claude and Gemini chats use your current switches from the next message you send.
- Codex chats take their connectors when the conversation starts. After you switch a connector on or off, start a new Codex conversation to use the change.
Workers and phone messages
Tasks a boss agent sends to its workers, and messages you send from your phone, use the same switched-on connectors as a message you type, limited to that session's project in the same way. See Boss and worker agents and Sessions and projects.
Limits
- Up to 20 built-in and custom connectors can be on in one chat at a time. Connected app accounts don't count toward this.
- Each connector uses room in the model's memory. The descriptions of a connector's tools are sent to the model with every message. A large connector, or many at once, leaves less room for your conversation. Connected apps take only a little room, however many you connect.
- Codex uses connector changes only in a new conversation.
Troubleshooting
If you see "This request was too big for …"
The message was more than the model can hold at once. When your connectors are the main cause, the message says how much room their tools took. Nothing has run out and your account is fine. To fix it:
- Switch off connectors you don't need right now, in the + menu under Connectors.
- Or switch the chat to a model with more room. See Connect an AI model.
- If the conversation itself is long, start a new chat.
If a connector you expect isn't in the Connectors panel
It may only run on another model; check the faint "N more works only in … chats" line. It may also be limited to other projects: check Available in on its card. A Claude account connector or Claude in Chrome needs a Claude chat signed in with your Claude account.
If a connector has an amber "!"
Its sign-in never finished. Click the ! and sign in again.
If the agent used the wrong account
Mention the one you mean with @ and send again.
If a Codex chat doesn't use a connector you just switched on
Start a new Codex conversation.