The Userflow MCP started out read-only: your AI tool could answer questions about your Userflow data, but it couldn't change anything. MCP write tools add eight tools to the same connection, so Claude, ChatGPT, and other connected AI tools can build the thing your answer points to—a flow, a segment, a chart, a dashboard—in the same conversation.
If you haven't connected the Userflow MCP yet, start with Connect Userflow to Claude, ChatGPT, and other AI tools with the Userflow MCP. If you're already connected, the write tools are there already. There's nothing to reconnect or reconfigure.
How writing works
Every write follows the same shape: you describe what you want, and your AI tool builds it and hands back a link so you can open it in Userflow and check it.
For flows, what comes back is an unpublished draft. It isn't live, and nobody in your product sees it until you publish it. Publishing is its own separate step that asks for your confirmation every time—it's never bundled into the instruction that created the draft.
Segments, charts, and dashboards work a little differently, because they don't have a draft state in Userflow. When your AI tool creates one, it exists in your account right away, the same as if you'd built it in the app yourself. Nothing about that reaches your end users on its own—a segment doesn't show anyone anything until a live flow targets it—but it's worth knowing the difference.
What you can create
Flows, guides, and campaigns
Describe a tour, tooltip, modal, announcement, banner, checklist, launcher, or resource center and your AI tool creates it as an unpublished draft, with steps and targeting built from what you asked for. Surveys and NPS work the same way, created as flows with question steps.
When a step needs to point at something on the page that's easier to click than to describe, your AI tool starts a Smartflow Builder draft instead and gives you a link to finish it by clicking through your own app.
You can also duplicate an existing flow into a new draft, which is a quick way to start a variant without rebuilding it.
Segments
Describe who you're targeting in plain language—"companies on the free plan who haven't upgraded after 14 days," "users who saw the onboarding flow but didn't complete it"—and get a real user or company segment built from Userflow's condition logic. You can reuse it in the same conversation to target an announcement, a banner, or a chart.
Charts and dashboards
Ask an analytical question and get a usage or funnel chart back, with the metrics, time period, and filters set up for you. Group related charts into a dashboard the same way.
Publishing a flow
Publishing and unpublishing a flow is a separate tool from creating or editing one. That's deliberate: it's the one write action that reaches your end users directly, so it always asks for its own explicit yes.
In practice this means a two-step exchange. You ask your AI tool to publish a flow, it comes back describing exactly what it's about to publish, and you confirm. If you'd rather publish from the Userflow app after reviewing the draft, that works too—the draft is a normal flow draft, and nothing about it is specific to the MCP.
Who can do what
Write tools run under your own Userflow permissions, using the same sign-in and role-based access as the rest of the product. If you can't publish a flow inside Userflow, your AI tool can't publish it for you either.
Every write is attributed to the teammate who triggered it. Nothing happens anonymously, and nothing crosses between accounts.
Because permissions come from your Userflow role, the way to change what someone can do through the MCP is to change their role in Userflow itself (see Managing Teams in Userflow), or to revoke their connection from the connection management screen in your Userflow settings.
Choosing which tools your AI tool can use
Your AI platform controls which tools it's allowed to call, not Userflow. In most tools you can open the connector's settings and switch individual tools on or off, or set them to ask you before running.
If you'd rather keep the MCP read-only for now, that's where to do it: turn the write tools off and leave the read tools on.
Working with write tools safely
Editing a segment that's already in use
A flow draft is versioned—when you edit one, the previous version is kept and you can get back to it from the Builder's version history. A segment isn't. If you ask your AI tool to change a segment's conditions, the change applies immediately, including to anything already using that segment.
Before changing a segment, it's worth asking what's currently targeting it. Creating a new segment is often safer than editing an existing one.
Making a lot of changes at once
A single open-ended instruction to change many things at once isn't a good use of these tools yet. Your AI tool carries out that kind of request as a long series of individual actions, and it's hard to review.
If you need to make the same change across many flows, break it into small batches, name the specific flows, and confirm each batch. If it's something you do regularly, a purpose-built skill is a better fit than a freeform prompt—talk to your Userflow contact.
Example prompts
"Which users signed up in the last 30 days and haven't finished onboarding? Create a segment for them, then draft an announcement about our new onboarding flow targeted to it."
"Show me the drop-off point in our Setup flow, then update that step to skip SSO users."
"Build a funnel chart for activation over the last 30 days and add it to my activation dashboard."
"Duplicate the Welcome Tour and call it Welcome Tour — Enterprise."
"Create an NPS survey for companies who've been customers for more than six months."
Frequently asked questions
Will my AI tool publish something without me approving it?
No. Creating or editing a flow always produces a draft. Publishing is a separate action that needs its own confirmation every time.
Do I need to reconnect or set anything up?
No. The write tools appear inside your existing Userflow MCP connection.
Who gets credited for something built through my AI tool?
You do. Every write runs under your own Userflow permissions and is attributed to you.
Can I keep using the MCP read-only?
Yes. Turn the write tools off in your AI platform's connector settings.
Does this use up my AI platform's credits or tokens?
That's determined by your AI platform, not Userflow. From Userflow's side an MCP call looks like any other API call. If you want to cap how much a job uses, you'd set that in your AI tool.
What happens if I hit Userflow's API rate limit partway through?
You'll get a message telling you the limit was reached and, where possible, when the action can be retried.
Can I have it delete things in bulk?
This isn't a good fit for an open-ended chat instruction. Anything irreversible is worth doing in small, named batches you can check as you go, or through a skill built for it.
Need help?
If something you built through your AI tool doesn't look right, or a write doesn't behave the way this article describes, reach out to your Userflow contact or post in the in-app help.