Give Your Dot Your Marketing: Connecting Marqeable’s MCP Server to OpenAI Dots
A dot is only as useful as what it can reach. OpenAI’s dots, launched September 29, 2026, connect to apps through plugins, and the working part of most plugins is an MCP server. Marqeable has one. This post covers how to connect it, which tools to let a dot run freely, which to hold for approval, and what the dot cannot do under any setting.
Updated September 30, 2026, after running this on our own dot. The connector worked on the first ask. The permission screens turned out to be different from what launch coverage described, and this post now reflects what is actually on the screen.
If you are new to dots, start with OpenAI dots for marketing.
What you need
| Requirement | Detail, as of September 2026 |
|---|---|
| A dot | Included with ChatGPT Pro and Business Premium in eligible markets |
| Developer mode | Needed for custom MCP connectors. Paid plans only, and in a workspace an admin may need to grant it |
| A Marqeable account | You sign in as yourself. The dot sees what your login can see |
| A desktop | Dots are created in the desktop app or a desktop browser |
Step 1: Add the connector in ChatGPT
Marqeable’s server address is:
https://app.marqeable.com/mcpIt is a remote server over HTTP with OAuth. No API key is stored anywhere.
- In ChatGPT, open Settings and turn on Developer mode. OpenAI’s Developer mode guide has the current location of the switch, which has moved between releases.
- Add a new connector, name it Marqeable and paste the address.
- Choose OAuth as the sign-in method and save.
- Sign in to Marqeable in the browser window that opens.
To confirm it works before involving a dot, open a normal chat, enable the connector and ask: “Run whoami and tell me which account I am signed in to.”
Step 2: There is no step 2
We expected a “connect apps to your dot” step for custom connectors. There is none. Dot onboarding says it plainly: “Your dot has access to your plugins on ChatGPT.” It offers to add Google email, calendar, contacts and Drive, and you can skip that. Anything already installed under Settings, Plugins, including a Developer mode connector, is available to the dot from its first message.
We tested it by asking a brand new dot: “Which apps can you use right now? Is Marqeable one of them? If so, run whoami.” It answered with the signed-in email and the current Marqeable account, and named the tool it called as mcp__codex_apps__marqeable_whoami. The same plugin serves ChatGPT, Codex and the dot.
Step 3: Set the plugin permission
The first control is not a rule at all. Under Settings, Plugins, each plugin has one permission level:
| Level | What it means, per ChatGPT’s own wording |
|---|---|
| Always ask | Ask before reading or making changes |
| Allow read-only tools | Read without asking, but ask before making changes |
| Allow low-risk tools (default) | Automatically approve low-risk tools, but sensitive tools may be denied |
| Allow all tools (marked “elevated risk”) | Use tools without asking |
Set Marqeable to Allow read-only tools. Marqeable’s server marks every tool as read-only or not, and this level reads that flag: reads run freely, anything that creates or changes something is supposed to ask.
One honest result from our test. With that level set, we asked the dot in chat to create one campaign draft, and it did so without a prompt. A dot treats an explicit request in the conversation as the approval, which is also the default in its rules (“Take action when you say so”). So the permission level protects you from the dot acting on its own initiative, not from what you ask for directly. That is the right way round, but worth knowing.
Step 4: Write the rules you need
Custom Rules live under Settings, Personalization, Custom rules. A rule is not a per-tool switch. It is a sentence you write, up to 400 characters, describing what the dot wants to do, plus one of four settings:
| Setting | What the dot does |
|---|---|
| Take action without asking | Acts on its own |
| Take action when you say so | Proceeds without asking if you explicitly request it |
| Ask before taking action | Asks for your approval first |
| Hand off to you | Asks you to do it yourself, and never does it on its own |
You can also ask the dot to propose a rule in chat, and it shows a confirmation form before saving. Here is how we would sort Marqeable’s tools into rules.
No rule needed: reads
These change nothing, and the read-only permission level already lets them run. Do not write rules for them, or the dot will ask you forty times a day.
| Tool | What it reads |
|---|---|
whoami, list_accounts | Who is signed in, and to which account |
get_business_context | Current focus, offers, audience and voice |
get_content_calendar | Calendar entries, planned content and suggested dates |
list_campaigns, get_campaign | Campaigns and their planned pieces |
search_content, get_content | Content pieces, briefs and current copy |
list_content_templates, search_image_library | Templates and existing images |
get_automation_overview, get_automation_plan | Automations, their steps and their numbers |
get_event_catalog, inspect_landing_pages | What an automation can start on, and page states |
check_account_setup, check_automation_readiness, validate_automation_plan | Checklists and dry runs |
get_brand_look | The account’s email look |
One rule: drafts and suggestions
Each of these creates or edits something a person reviews in Marqeable. Nothing reaches a customer. A single rule covers them: “When your dot wants to create or edit a draft, suggestion or brief in Marqeable,” set to Ask before taking action while you are learning its judgment, then Take action without asking once its output has stopped surprising you.
| Tool | What it creates |
|---|---|
create_campaign_draft | A campaign in draft |
create_content_draft | A content piece in draft |
create_automation_draft, update_automation_draft | An automation draft, or a proposed revision a person publishes |
create_automation_step_content | Draft content for an automation’s empty steps |
add_content_suggestion | A suggested edit the owner accepts or dismisses |
update_content_brief | Changes to a draft’s brief |
request_content_changes | A change request at a build checkpoint |
Ask before taking action: direct changes
These take effect without a review step in Marqeable, so put the review in the dot’s rules: “When your dot wants to change the marketing calendar, brand look, templates or approve a content build in Marqeable,” set to Ask before taking action.
| Tool | Why it needs a yes |
|---|---|
add_calendar_entry, update_calendar_entry | Calendar changes feed the next plan directly |
adopt_suggested_date, dismiss_suggested_date | Same, for suggested dates |
approve_content_build | Moves a build forward and can generate images |
create_automation_trigger_page | Builds a landing page for an automation |
save_content_as_template, set_default_content_template | Changes what future pieces start from |
update_brand_look, reset_brand_look | Changes the look of every email |
Hand off to you
| Tool | Why |
|---|---|
set_feature_enabled | Turns account features on or off. A person’s decision |
switch_account | Should only happen when you explicitly ask for it |
| Signing in to Marqeable in the dot’s browser | Never. See the first test below |
The rule: “When your dot wants to switch Marqeable accounts, change account features, or sign in to Marqeable’s website,” set to Hand off to you.
What the dot cannot do, under any rule
There is no tool to send a message, publish content, launch a campaign or turn on an automation. There is no delete tool. A rule cannot allow an action that has no tool.
That is the point of the design. Rules are a setting, and settings drift. The tool list does not. The longer argument is in the draft-only rule.
One honest caveat: not everything is a draft. The calendar, brand look and template tools above make real changes. None of them reaches a customer, but they are why the approval group exists.
A first test
Once connected, give the dot a small job with a visible result.
You: Check my marketing calendar for the next four weeks against my current focus in Marqeable. If there is a gap, draft one campaign to fill it and send me the link. Do not change the calendar.
Expected result: A handful of reads, one campaign draft, and a link to review it in Marqeable.
Here is what our run looked like. The dot called whoami, get_business_context, get_content_calendar, list_campaigns and get_automation_overview, some in parallel, and reported one calendar entry with no campaign linked and three empty weeks. Asked to fill the gap, it created one campaign draft and returned the review link. The draft sat in Marqeable in Draft status with nothing built, sent or scheduled.
We then had the dot schedule a one-off task for five minutes later: whoami, then count the calendar entries for the next 30 days, read only. It ran about five minutes behind its scheduled time, used the connector unattended, and posted the account, the date window and the counts into the dot’s chat. So the connector works in background runs, not only in conversation.
Two things to know from that run.
First, we could not find an activity log. The dot’s side panel showed “No recent activity” and “No outputs yet” after the work was done, and the dot itself could not point us to one. What worked was asking: “List the exact Marqeable tool calls you made, in order.” It answered precisely, down to the arguments. Until a log turns up, that question is the audit.
Second, a dot will look for another way in when a tool does not do exactly what you asked. Our first draft request specified one email and one text. Marqeable’s campaign builder chooses the content mix itself, so the dot announced it was “checking the manual route” in its cloud browser, which meant signing in to Marqeable’s website. We told it to stop and use only the connector, and it did. Never give a dot your password to work around a tool. Write the hand-off rule above, and phrase requests in terms of what the tool can do.
Troubleshooting
| What you see | What to try |
|---|---|
| No Developer mode switch | Check your plan, or ask your workspace admin for custom connector permission |
| Sign-in succeeds but no tools appear | Remove and re-add the connector, then start a new chat |
| Dot asks before every read | Set the Marqeable plugin permission to Allow read-only tools |
| Dot wants to sign in to Marqeable in its browser | Tell it to use only the connector. Never hand over a password |
| Wrong account | Ask the dot to run whoami. Switch accounts yourself, deliberately |
From here, five dot workflows for a marketing team of one puts the connection to work. The same server in a terminal is covered in Codex MCP for marketing.
Frequently asked questions
Can an OpenAI dot use a custom MCP server?
Yes. A connector added through Developer mode in ChatGPT is available to your dot with no extra step, because a dot has access to the plugins installed on your ChatGPT account. We confirmed it on our own dot on September 30, 2026: it ran Marqeable’s whoami on the first ask, read the calendar and campaigns, created a campaign draft through the connector, and used the connector again inside a scheduled task that ran while nobody was in the chat.
What is the address of Marqeable’s MCP server?
https://app.marqeable.com/mcp. It is a remote server over HTTP with OAuth. You sign in with your normal Marqeable login, and the assistant sees exactly what your account can see.
Can a dot send a campaign through Marqeable?
No. Marqeable’s connector has no send, publish or launch tool. A dot can create campaign, content and automation drafts, and a person reviews and launches them in Marqeable.
Which Marqeable tools should a dot need approval for?
Anything that changes account-level settings: the brand look, the default template, turning features on or off, and switching accounts. Also approving a content build, because that step can generate images. Reads can run freely under the Allow read-only tools permission, and draft creation is a reasonable thing to allow once you trust the dot’s judgment.
The bottom line
Connecting a dot to your marketing is three decisions: which server, which tools run freely, and which tools wait for a yes. With Marqeable the first is one address, the second is one permission level, and the third is two short rules. Sending is not a decision you have to make, because the tool does not exist.
See what the drafts become: Marqeable’s agents build and run your campaigns, run behavior-triggered automations across text and email, answer and qualify every visitor with AI website chat, and tie revenue to the exact message with attribution.
Marqeable runs your campaigns, answers every visitor, text, and email in seconds, and turns them into booked jobs and meetings - even at 9pm on a Saturday. We’re in private beta with a small early cohort. Get early access
