Connect Claude or ChatGPT to omotto

A connected agent is your own AI helper authorized to work in omotto. An AI agent is an assistant like Claude (made by Anthropic) or ChatGPT (made by OpenAI) that can use outside tools when you ask it to. Once connected, you can ask it in plain words, for example: “Build an automation that notifies me when a refund is created”, and it works in omotto for you.

How it works

The connection uses MCP (Model Context Protocol), a common standard that lets AI agents talk to other apps. You do not need to know how it works.

This is the place for building automations and Custom API connections (connections to your own system; API means application programming interface) with AI help. The Coach inside omotto does not build those itself. When you ask it, it points you here.

What a connected agent can do

With read access, your agent can:

Read your workspace facts. It can see team members and roles, your business profile and support hours, your plan’s features and limits, business days, setup progress, explanations of settings, handoff alert recipients, label rule attributes, and which integrations can be connected.
Search your catalog and read its notes. It can find products in your live store or website catalog, read best-seller rankings when reporting is enabled and your store allows reports, and read the catalog notes your bot uses. A website catalog does not prove current stock or supply best-seller rankings.

Changing Who is this for on an existing skill, endpoint, Expert, File rule or group-sharing capability opens an in-app access step. Applying or reverting an audience change follows the same rule. Creating a new item with an audience remains direct. Existing verification/proof settings remain directly editable with the same permissions and validation; changing a proof setting does not add an in-app access step.

With Make changes allowed, your agent can:

Review the Coach’s pending suggestions. Your connected agent can see suggestions you are allowed to review, then apply or dismiss them for you. Applied changes are recorded and have the same undo options as in the app. If a change has no undo, the agent is told before applying it.
Build and edit automations. It can start from an idea or a blank draft, choose the moment, add waits and condition checks, arrange actions, and turn an automation off.
Build Custom API connections. It can set up a connection to another system and its endpoints, or import a ready-made setup, so your bot can look things up or act in that system. It can read the connection’s readiness and tell you what is missing. Setup is not complete until the connection shows Working; review the endpoints, enable the ones you need, and test them.
Prepare user sources. It can draft a source that checks for accounts in your business system. It starts switched off, with typed-detail lookups off too. When the agent reads a source’s setup, unreadable saved access details make it unavailable even if the source is still switched on; the agent receives the reason, not the secret.
Set up connections to Zapier, Make and n8n. For new automatic WhatsApp template sends, ask it to build an automation with a Send a template action. It no longer creates old-style message triggers.
Maintain existing message triggers. It can edit their mappings, preview them and send a test to your own verified phone. Reusing a saved customer example for a test needs conversation access; otherwise the agent supplies its own example. It can turn a trigger on once its sending setup is ready. That starts real customer sends immediately. The agent is told to strongly recommend that you review it now and gives you a link.
Draft WhatsApp templates. You still send them for approval yourself.
Read and change many bot settings that you can change yourself. This includes turning the bot on or off for the business or one line, related-customer sharing, Expert access settings, who may approve an action, verification strength, and switching File rules on or requiring a verified customer. These changes use your permissions, are recorded under your name and can be reverted from Recent changes. Turning the bot on checks setup readiness. If something is missing, the agent tells you what it is; you can provide a reason to go live anyway. Turning it off needs no reason. With Shopify connected, the agent can discover and change individual order-change policy settings, including eligibility, allowed ways to act, verification and approval. It can also create and delete custom order-change types if you have permission to edit these settings; creating or deleting a type happens directly and can’t be undone from Recent changes. Built-in types can’t be deleted.

Once you have reviewed your business name and description, you can ask the connected agent to confirm them on your behalf. Both a name and a description must be filled in. If the text has changed since the agent read it, it must read it again before confirming. If the details are already confirmed or still incomplete, nothing changes. Editing either later removes the confirmation; changing the text does not confirm the new version.

Read and change the business calendar, including its market, working week and specific closed/open dates, with your Workspace settings permissions.
Set up Google Calendar. It can create a meeting type or a bookable calendar, and set a calendar’s location and the meeting types it offers, with the permission to edit the Google Calendar connection. It picks locations and meeting types by name from lists omotto gives it. A repeated identical request returns the first result instead of creating a copy. Undoing a create in Recent changes switches it off rather than deleting it, and only while nothing about it has changed. See Connect Google Calendar.
Ask for a check of a specific assistant reply, when conversation access is also allowed. The verdict and any suggested fixes appear in a Coach conversation for you to review and approve. Your agent can read the verdict and the proposed cards, and ask follow-up questions. Asking again about the same reply returns the existing check, rather than starting another paid analysis. A follow-up can request a fresh check after it finishes. A paid analysis that has started cannot be undone.
Read and change skill instructions. For any skill whose instructions the Coach can edit, it can read the instructions actually in use and change them, with your existing settings permissions.
Read and change many bot settings that you can change yourself. A few settings can only be changed in the app. With Shopify connected, the agent can discover and change individual order-change policy settings, including eligibility, allowed ways to act, verification and approval. Creating or deleting a custom order-change type uses a separate reviewed proposal through the Coach; it is not part of these connected-agent settings.
Change location details and availability. It can update an existing location’s name, address or service phone, close a location or one calendar for a date, and change dated or weekly hours. A calendar-only day off leaves the location and other calendars open. A dated exception keeps following the location’s weekly hours; a calendar’s weekly-hours edit gives it its own hours. Existing appointments are not cancelled. You can ask the agent to undo an eligible change from the recent-changes list; a newer change to an affected schedule can prevent undo.

For “close tomorrow”, tell the agent which location you mean. For one person’s day off, name their calendar. The agent reads omotto’s local date and current hours before changing them.

With read access, your agent can also:

Read websites for you. It can search and read the documentation sites you configured for the bot and made available to your team. It can also read a public web page for setup research. These use the same daily limits as the bot and the Coach, respectively.

Your connected agent can read the accepted fields and their help before making a change. For Custom API operations, it can also see where inputs are sent, the allowed kinds of input, and values omotto can fill in itself. Each value says whether it is system-held, verified or as given. It can read why an operation is not ready to use. A misspelled field or a list or object sent in the wrong shape is refused with the field named, rather than silently ignored.

A connected agent can also read omotto’s current time.

Searching and reading your knowledge

With access to the Coach, you can ask a connected agent to search your knowledge, open a document or a section, read spreadsheet cells, or browse the available documents. Large answers arrive in parts so the agent can keep reading. A source whose access has been removed is not served as knowledge, even if its entry still appears in the document list in the app.

It can also find the skills, documents, File rules and saved business memories that an existing teaching, editing or deletion action can use. This lookup does not change anything.

Reading the original text used to edit a document requires both permission to view and permission to edit knowledge. This is separate from reading the text the bot currently serves. The original keeps time-limited information’s markup so an edit does not silently remove it.

With Coach access, the agent can look up past knowledge changes, including the old and new wording. Versions of customer-specific memories are left out unless you granted conversation access. Without that access, the result says that customer-memory history was omitted.

Find and read conversations, with conversation access allowed. Search by customer, your own involvement, words or dates. Read the customer chat alongside its internal channel, follow omotto links, and see the saved sources behind a bot answer. Only conversations you may open are included.

Teach and edit knowledge

For a custom skill, the connected agent can read, set or clear its Until date. Expired skills remain visible but the bot no longer uses them or their attached knowledge. Undo restores the previous date, including an already-expired date.

Your connected agent can create, edit and delete business memories, custom skills and text-based knowledge documents, using the same saving rules as the Coach. It can also switch custom skills on or off. It acts with your permissions. Personal customer memories are not part of these knowledge-object actions.

Changes are recorded, including changes applied from Coach cards. Ask the agent to undo a change, or use the recent-changes page below. Undo refuses if someone has edited the record since. Restoring a deleted memory brings back the same memory; restoring a deleted skill or generated text document creates a replacement with a new reference. A skill that owns documents or web sources cannot be deleted until those sources are dealt with.

If you try to re-create a removed memory about a specific product or other business item, the agent refuses. Ask it to undo the original deletion instead; this restores the same memory. A file finishing its safety scan does not count as someone editing the document and does not prevent undo. If the edited file is still being checked, wait for the check to finish and ask the agent to undo again. Some files contain text that cannot be restored exactly; in that case, undo is unavailable, so keep the original file.

Deleting an uploaded-file knowledge document permanently deletes its file. That deletion cannot be undone. The agent is warned before it calls delete. Provider-synced document bodies must be changed at their original source.

Send files, add sources and run imports

A connected agent can upload a file for you, either directly or through an upload link. Files are private to the member and workspace that uploaded them. They are scanned before use. Temporary uploads are deleted after 30 days; material already saved from them keeps its own copy.

The agent can use files for File rules or knowledge documents, add Web sources, and add Google Drive files that you have already selected for omotto in the file picker. A file rule sends a file; a knowledge document lets the bot answer from its contents.

For a knowledge import, the agent can start from a website or a supported conversation export, check progress, confirm who is staff, inspect the complete review and make individual decisions. Conversation exports, customer notes and customer links require Read conversations. If a review contains those items, that permission is needed to apply or discard the whole batch too.

Individual decisions can be undone while the review stays open, unless someone has changed them since. Customer-text edits cannot be undone from change history. Applying or rejecting the whole review is final. Applying accepts items without an explicit decision too. Starting a website import also fills your business website if it was blank. Ask the agent to show you the review before it finishes.

What starts off, and how to make it live

New Custom API connections start on if created with a complete host address. Drafts with no host yet start off. An omotto preset follows the hosted-connection default. Endpoints and external actions start off, so nothing runs through them until they are switched on. Required sign-in details must also be present.

The connected agent can switch Custom API endpoints and Zapier, Make or n8n actions on with your permissions. It clearly says when something is not yet operational, recommends reviewing each item first, and supplies links. You can ask it to switch the item on or do it yourself in omotto. Switching on an endpoint is not enough if its parent connection is off. The agent reports the connection state and whether saved sign-in details were detected. This is not proof that those details match the current authentication method or that the provider accepts them. Check Authentication and its required fields in omotto before use.

For Custom API endpoints that change something in another system, Ask a teammate before each call is a per-endpoint choice. It is off for endpoints created from scratch; imported presets and templates can carry approval on. Existing write endpoints kept it on. The agent can change the choice with your permissions. Read the Custom API guide before changing it; permission to try an action is not proof that it succeeded.

Editing a connection or operation

An authorized connected agent can raise or lower customer-verification and user-source requirements directly, with the same permissions and validation as your app editor. It can also edit methods and real-action classification while an endpoint is on. Secret entry and real tests keep their existing in-app steps; editing does not test or send anything.

Changing a connection’s base address (even only its path), token address, authentication method, header/query name or client-authentication style removes its saved sign-in credentials, cached token, secret request headers and fixed secret input values on its endpoints. Input names and positions stay. Affected endpoints switch off and stay off until you re-enter those values in omotto and switch them back on. A relative path change on one endpoint does not trigger this removal. In the app, choose Save connection changes and review Remove stored secrets and save? before Save and remove secrets; Cancel keeps the saved values. After saving, the result lists what was removed and links to re-enter it. A connected agent warns you beforehand with the exact secret fields and endpoints; it applies the change directly, without a separate approval step. A receiving-address change on a Zapier, Make or n8n action removes an operator-entered secret header; a secret generated by omotto stays. The agent reports removal and gives you a re-entry link. If an address itself contains a removed secret and would otherwise become visible, that address is cleared too and the connection switches off. The warning and result name the address fields to re-enter before switching it back on.

A connected agent can preview automations, show which request values would be sent, and run one live test of an eligible web-request or external-action step. A live test really contacts the destination and can make a real change. It never runs the automation’s WhatsApp-template or team-notification steps. The agent can read the latest test outcome and recent normal-run failures without receiving response bodies.

For Shopify automations, the agent can discover order and item values, separate shipping and billing addresses, buyer name and tracking. Ask for the amount with currency when filling a message variable; the numeric amount is a separate choice for other systems. Missing store details remain unavailable. See Shopify values and Preview privacy.

New automation drafts start off. You can ask the agent to enable them, including automations that send conversation content. It recommends reviewing the editor link first. Enabling and edits of a live automation report live now, the destinations and the conversation content being sent. Live edits take effect immediately, just like saving in omotto. Review privacy and the receiving workflow before enabling or editing.

What stays in the omotto app

Some steps are always done by a person, inside omotto. The agent prepares the work, gives you a link, and you finish it:

Setting up an idea that needs requests from your system. The agent links you to the idea in omotto without creating an automation or incoming address for it. It does not receive the source’s secret. Finish that setup in the app.
Turning a user source on. Review its setup in omotto, then enable it yourself.
Changing what a live user source accepts. Changing its response mapping or the declaration that the lookup phone/email belongs to the account holder stays in the app while the source is on.
Looking up details people type. Only you can turn on Also look up details people type (unproven). The agent cannot run live user lookups or Try it; you test the source in omotto.
Deleting an automation.
Passwords and keys for other services. You enter these in omotto, not through the agent. For a Custom API connection, save every credential needed by the selected authentication method in Manage connection. The agent can report the missing fields and link you there. The bot is not offered that connection’s operations until those credentials are saved, even if the connection and endpoints are on. The agent can receive secret codes that omotto generates for incoming connections, such as Zapier, Make, n8n or message triggers. Treat those codes like passwords and share them only with systems you trust.
Messaging consent. Confirming that customers agreed to marketing messages is done by you in the app. See messaging consent.

When the agent hits one of these steps, it tells you and gives you the link to the right page.

Where to find it

Open Settings → People & access → Connected agents.

Set it up

You need an account with the AI agent. Claude and ChatGPT decide which of their plans can use connections like omotto. Check their current help pages.
The agent acts as you. It signs in as you, in one workspace you pick, with exactly your permissions. If you cannot do something in omotto, your agent cannot either.

Step 1: choose your agent

Open the screen listed above. Under Choose your agent, pick Claude, ChatGPT or Another agent. The steps below the choice change to match it. A Connected badge on Claude or ChatGPT means you already have a usable connection with change access.

Step 2: add omotto to your agent

Follow the numbered steps shown for the agent you picked. The AI companies rename their menus from time to time. If a name below does not match, look for “connectors” in their help.

Claude (on the web or the desktop app)

omotto is in Claude’s connector directory, the list of connectors Anthropic (the company behind Claude) has reviewed. You add it from there and do not need any address.

Click Open omotto in Claude. It opens omotto in Claude’s connector directory in a new tab. If it does not open, go to Customize, then Connectors, click +, and search for omotto.
Click Connect to Claude.
Sign in to omotto and approve the connection (Step 3 below).

On Claude Team and Enterprise plans, an admin adds omotto for the team first: Organization settings, then Connectors, then Browse connectors. They pick omotto and click Add to your team. Members then find it in Customize, then Connectors, and connect it themselves.

To use it in a chat, click + at the bottom left of the chat box, choose Connectors, and turn omotto on.

ChatGPT (on the web)

Click Open ChatGPT. It opens ChatGPT’s Plugins page in a new tab.
Click Add, then Add custom MCP server.
Type a name, such as “omotto”. Under Connection, paste omotto’s server address and choose OAuth for sign-in. The address is shown in this step on the Connected agents screen, with a Copy button. It ends in /mcp, for example https://app.omotto.ai/mcp.
Read the warning and tick I understand and want to continue. Click Create as a plugin.
Install the plugin and sign in (Step 3 below).

The server address is not a password. Nothing can connect without you approving it. If copying fails, select the address and copy it manually.

To use it in a chat, type @ in the message box and pick omotto.

Another agent

Any AI agent that can connect to outside tools over MCP (the standard way AI agents connect to tools) can use omotto.

In your agent, look for custom connector or remote MCP server.
Paste omotto’s server address, shown in this step with a Copy button. If the agent asks how to sign in, choose OAuth.
Sign in to omotto and approve the connection (Step 3 below).

Step 3: approve it in omotto

The Approve it in omotto section reminds you to choose this business and allow changes. Your AI agent opens an omotto page. Sign in if asked. The page asks Connect …? with the agent’s name.

Under Organization it may act in, pick the workspace.
Under It will be able to, the agent can always read data in that workspace, as you. Tick the extra boxes you want:
- Make changes in that organization, as you lets it create and edit things. You need this one for building automations and Custom API connections.
- Read conversations, audit findings and your reply analyses, as you lets it look at customer conversations. Tick it only if you want that.
Click Allow.

The agent’s name is not verified by omotto. Only connect apps you trust and set up yourself.

Reading conversations is its own separate approval: ticking Make changes does not include it, and it is not given unless you tick it.

Read-only access lets the agent look, but not build. If it is already connected read-only, reconnect and grant Make changes in that organization, as you. A connection to another workspace does not let it build in this one.

Back on Connected agents, Things you can ask has example requests. Your connections shows each connection’s workspace, access and last use.

What happens when…

You want to review or undo a change

Every change your agent makes is recorded. On the screen listed above, open Recent changes (if your role allows it). You see changes made by connected agents and by the Coach, and what connected agents read.

When the agent changes a style setting that has your own text, omotto tells it that your text will be replaced, and you can undo the change in Recent changes.

Click Revert to undo a change from the last 30 days. Reverting a newly created automation or connection turns it off. Reverting a newly created WhatsApp template draft deletes the draft, rather than keeping it switched off. Reverting a newly created Google Calendar meeting type or calendar turns it off and deletes nothing; it is refused if the item was edited after it was created. Some changes cannot be reverted here; the page tells you, and you change them on their own settings page instead.

You want to disconnect an agent

Open the Connected agents screen listed above.
Find the agent and click Disconnect.
Confirm. The agent loses access to that workspace immediately. This ends the whole approval you gave it for that workspace, including the extra boxes you ticked in Step 3. To use it again, connect it again and approve it again.

You can also see and remove your connections from My account → Connected agents. There, each connection shows its workspace, the access you granted, when you first approved it and when it was last used. Revoke access stops that agent in that workspace only; your other workspaces and other agents are not affected.

If you leave a workspace or lose access, your agent loses access too.

Troubleshooting

The agent says it cannot do something. Check your permissions in omotto. The agent can do only what you can.
The agent cannot connect. In Claude, add omotto from the connector directory, not as a custom connector. For ChatGPT and other agents, check the server address ends in /mcp, and that you signed in to the right omotto account. For current external menu names, see the official Claude connectors guide or ChatGPT connection guide.
A connection that worked before now asks to be approved again, or stops working. After a recent omotto update, every existing agent connection has to be reconnected once. Disconnect it if it is still listed, then add it to your agent again and approve it in Step 3. This is a one-time step. Turning connections off and on again, or leaving and rejoining a workspace, does not bring an old approval back.
The agent can read but cannot build. Reconnect, choose the workspace where you want it to work, and allow changes in Step 3.
Still stuck? Contact omotto support.

When an automation save is refused because a Custom API target has setup problems, the connected agent receives the same addressed explanation as the editor, not just a generic failure. Repair the endpoint definition first.

Sending values omotto already knows

A connected agent can prepare Custom API inputs using Value from omotto, such as a conversation id or verified customer email. Adding, removing or changing these server-held inputs applies directly, just like editing in the app, including when creating or importing an operation. The result names the values and their destination without asking for approval. As-typed contact details are labelled separately and do not prove who owns them.