Connectors

Connectors lets agents act as accounts at outside services, such as an X account or an SMS provider, through tools you define as data rather than code (module connectors). You add an account once, give it to the agents that may use it, and every call they make is checked against that grant before anything is sent. The X screen lets you use your X accounts yourself, without an agent.

In the sidebar under Connections → Connectors and Connections → X. Needs: Agents. Handles secrets: account keys are stored here, so anything an agent does in this module other than reading goes to a person for approval first, unless a person has chosen Always allow for that exact tool and agent.

Connectors

The page has seven tabs.

• Accounts: one account at one outside service per row: its label, its service, whose it is, its health, and whether agents may read only or read and act. A key is typed here once and never shown again, to anyone. Above the list, the Services card offers services that are ready to set up, such as X. An account that is managed in Integrations says Manage in Integrations.
• Tools: the connector tools defined here, each one https request to one service, with its service, status (draft, published or off) and version.
• Calls: every call made through an account, newest first. Addresses are not kept, so a key in one never is.
• Posts: every post agents made to a social channel (X, Threads, Instagram, YouTube, LinkedIn): where, as which account, its link, and whether the platform showed it or it was deleted.
• Costs: what connector calls cost over a range of days, by agent, by account, by person, by service and by sign-in app, with the newest calls and a CSV download. It also shows each sign-in app's daily platform allowance, such as YouTube's uploads. Costs are estimates from each tool's stated price. A call with no price is counted as unpriced, never as free.
• Requests: accounts and outside MCP servers that agents asked for.
• Sign-in apps: a service's sign-in settings, so its accounts can be connected by signing in rather than pasting a key. This includes your own app, paid from your own credits.

Accounts and who may use them

An account is either Mine (only you see it and give it out) or The organisation's (only someone who may manage connectors can add one). An agent acts only as the accounts a person gave it. Each account also lists the hosts its key may be sent to. A call to any other host is refused, and an account with no hosts refuses every call.

Tools defined as data

A connector tool is one https request: a method, an address with its host written out, and {{input.x}} placeholders for what the agent fills in. You draft it, try it against a real account (the request is shown with the key masked), and publish it. Published, it is offered as connectors.<name>, for example connectors.x_post, to every agent and key in this workspace, under the same permission and approval checks as any other tool. Tools have versions and can be switched off with a reason.

Because these tools are made at run time from what is stored, they are not a fixed list. They appear in the tool list once they are published here. Tools imported from the Connector hub appear here too.

The X catalogue

Setting up X adds its tools in one go, for reading, posting, likes, reposts, follows, bookmarks, lists, hiding replies, muting, blocking and direct messages, for example x_me, x_search, x_post, x_reply, x_like, x_follow, x_bookmark and x_dm_send. Each is offered as connectors.<name>. When an agent calls any of them that is not a read, such as posting, liking, following or sending a message, the call waits for a person. Pressing Always allow on the approval card lets that agent make that one call (say, likes) without asking from then on, while posts and messages still ask. X charges per call, and each tool states its own price, which the Costs tab uses. Direct-message tools work only on an account that granted direct messages when it signed in.

X

Everything you do on X yourself, as the X accounts you own or that are shared with you. Pressing a button is the approval, and every act is recorded under your name. Choose the account at the top. The tabs are:

• Write: a post or a thread, with pictures, a GIF, a video or a poll, who may reply, the Made with AI and paid-partnership labels, and a community. Its price by X's list is shown before you post. Post now, Save draft to keep it, or schedule it for later.
• Drafts & scheduled: what is kept or scheduled. Waiting for your OK lists each post, reply, thread or schedule an agent asked to make. When something waits, the tab reads "Drafts & scheduled · N waiting". A scheduled post goes out by itself within a minute of its time, as you, with nobody asked again.
• Posts: what this account posted through Max, and its latest posts on X, with Delete.
• Inbox: mentions to reply to, like and repost, and direct messages for the account's Managers.
• Activity: one plain line per act, by a person or an agent, with what it acted on, how it ended and what it cost. Never a message's text.
• All accounts: mentions and Activity across every X account you can see.
• Share: the owner gives the account to a person, a team or the whole organisation as Viewer, Poster or Manager.

The page also shows the account's numbers (followers, posts and their views and likes), lists, people, spaces, search and its kept archive.

What you can do

Set up a service and connect your account

1. On Accounts, find the service on the Services card.
2. Someone who may manage connectors presses Set up X once and pastes the client id and secret that the service gave this deployment. The sign-in app is added and its tools are drafted.
3. Anyone who may use connectors then presses Connect an X account. Each press signs in as whichever account is signed in at the service in that browser, and names the account after it.
4. To connect a second account, switch accounts at the service first, then connect again.

When a service has more to offer, the card shows Set up again. Each account's owner then signs in again once to grant the new permissions.

Use your own developer app

A person can add their own app for a service with Add my own X app. Accounts connected through it spend that person's credits, not the workspace's, and nobody else can use it. Use my own app on an account moves it to your own app. On a workspace app, Make each person bring their own stops others connecting through it; accounts already on it keep working.

Add an account with a key

1. On Accounts, press Add account.
2. Choose the Service, or Another service… and type its short key.
3. Give it a Label, such as @acme, which agents and approval cards will show.
4. List the Hosts its key may be sent to.
5. Choose Whose, and what Agents may do: Only read through it, or Read and act — post, send, change.
6. Choose how it Signs in with: a bearer token, a key in a header, a user name and password, a key in the address, OAuth 1.0a keys, or signing in at the service. Paste the key and save.

Give an account to an agent

1. Press Give to an agent on the account and pick the Agent.
2. Press Who can act as this to see every grant.
3. Test makes one read call as the account to check it still works. Make read only, Switch off and Replace key are on the same row.

Define and publish a tool

1. On Tools, press New tool.
2. Give it a Name (offered as connectors.<name>), a Service and What it does, then its inputs, its request and what to pick from the answer, its class, a Cost per call and an Undo tool if there is one. Press Save draft.
3. Press Try, choose the account to act As, enter the Inputs (JSON), and press Send.
4. Press Publish. To stop a tool, press Switch off and say Why.

Answer an agent's request

1. Add the account the agent asked for on the Accounts tab, and give it to the agent.
2. On Requests, press Done, pick the account you added, and press Mark done.
3. Or press Decline and say why.

Who can do what

Permission	Label	Lets you
connectors:read	See handles and connector tools	Open the page and see accounts, tools, calls, posts and costs.
connectors:use	Call connector tools	Call tools, add and connect your own accounts, use the X page, and give your accounts to agents.
connectors:manage	Define tools, organisation handles and sign-in apps	Define and publish tools, add the organisation's accounts and sign-in apps, set up services and decide requests.
connectors:author	Let an agent add tools and ask for handles	Lets an agent add a tool for itself and ask a person for an account.

A key is always typed by a person signed in as themselves, or taken from a stored secret by name, never returned. On X, what you may do on a shared account follows its share level. See People, roles and permissions and Sharing and visibility.

What happens on its own

Job	When	What it does
renew-and-warn	Every day at 03:00 UTC	Renews social accounts close to lapsing, warns owners 7 days before a sign-in that cannot be renewed ends, deletes old media links, marks accounts whose owner has left, and removes Activity lines older than 400 days.
crosspost-sweep	Every minute	Resumes a cross-post that stopped part way.
account-days	Every day at 02:40 UTC	Records each X account's numbers for the day, because X keeps no history.
x-autocheck	Every 5 minutes	Checks X accounts that have automatic checks switched on for new mentions and direct messages.
feed-sweep	Every 5 minutes	Handles X feed events left over, and hourly checks each feed's health.
x-scheduled	Every minute	Sends scheduled X posts and threads whose time has come, each once.
x-keep-media	Every 5 minutes	Copies pictures and videos read from X into Max's storage.

Needs

The Agents module.

What agents can do

An agent can see the accounts it may act as, the tools and their costs, add a tool it needs on an account it already has, and ask a person for an account it lacks. Connectors handles secrets, so every act by an agent here other than reading goes to a person, unless an Always allow rule names that tool for that agent. Some acts always ask, every time. See Agents and approvals.

Read only: connectors.handles, connectors.agentHandles, connectors.tools, connectors.costs, connectors.channels

Tools and requests: connectors.add_tool, connectors.request_handle, connectors.tool_off

Accounts, always ask: connectors.set_up_preset, connectors.add_account, connectors.give_account, connectors.replace_key, connectors.register_account, connectors.renew_key

Accounts, change: connectors.rename_account, connectors.switch_off_account, connectors.take_account (the last two only narrow what can be done)

Posting: connectors.crosspost posts one item to several social channels under one approval.

X: connectors.x_activity, connectors.x_drafts, connectors.x_numbers, connectors.x_saved (read only); connectors.x_draft_save, connectors.x_unschedule (change); connectors.x_thread, connectors.x_schedule, connectors.x_many (act on X, need approval)

• connectors.add_tool publishes a tool for the calling agent alone, on an account it already has. Its first use, and the first use of every new version, always goes to a person. It never takes a key. An agent can add at most five tools an hour.
• connectors.request_handle asks a person for an outside account or MCP server. Nothing becomes usable until a person adds it and gives it to the agent.
• connectors.add_account and connectors.replace_key take the name of a stored secret, never the key.
• connectors.x_saved is free and sends nothing to X: it reads what Max already kept.

Beyond these, every published connector tool, such as connectors.x_post, is a tool agents can call, as the account they were given.

Over MCP the same tools are named connectors__handles and so on. See API keys and MCP access.

Related

• Connector hub
• Agents: who acts as these accounts.
• Approvals: where calls that act wait for a person.
• MCP: outside MCP servers, given to agents on the Agents page.
• Agents and approvals
• API keys and MCP access