Bubble MCP
Bubble's MCP server connects AI tools like Claude, Claude Code, and Codex to your Bubble account so they can build and edit Bubble apps. Covers setup, available tools, and limitations.
Bubble's MCP server lets you connect an AI tool like Claude, Claude Code, or Codex to your Bubble account. Once it's connected, you can ask the AI tool to create a new app or make changes to an existing one, and it builds directly in Bubble. The result is a regular Bubble app that you can open, review, and edit in the visual editor like any other. The Bubble MCP supports building both web apps and native mobile apps.
This article explains what MCP is, why you might use it, how it differs from Bubble AI, and how to connect your AI tool.
What is MCP?
MCP stands for Model Context Protocol. It's an open standard, introduced by Anthropic in 2024, that gives AI tools a common way to connect to other software. Without a shared standard, every AI tool would need its own custom integration for every service it wanted to work with. MCP replaces that with one protocol that any compatible tool can use.
A helpful way to picture it is a USB-C port. Just as USB-C lets you plug many different devices into one kind of socket, MCP lets many different AI tools plug into the same services. Bubble provides an MCP server, which means AI tools that support MCP can connect to Bubble and work with your apps.
When people talk about "an MCP," they usually mean an MCP server: the connection point a service like Bubble provides for AI tools to plug into.
Connecting an external AI agent to Bubble
Every tool follows the same basic pattern: you add Bubble's MCP server to your AI agent, then sign in to your Bubble account in the browser to give the tool access.
Claude
Claude Code
ChatGPT
Codex
VS Code and GitHub Copilot
Why connect an external AI agent to Bubble?
Bubble AI and the MCP server are two ways to build with AI on Bubble, and both build the same kind of app. The difference is where you work and which AI does the work. Here's why you might choose to connect an outside AI tool.
Work in the agent you already use
If you already spend your day in Claude Code, Codex, or another AI agent, you can build and edit your Bubble apps from there, without switching back and forth. Your conversations and your way of working stay where they are.
Choose your AI model
Most AI tools let you pick from one or more AI models. With the MCP server, you decide which model builds your app, and you can switch whenever a different model suits you better.
More variety in design
Every AI agent and model tends to build in its own recognizable way, and that includes design. Given the same prompt, two agents can make different choices about layout, spacing, color, and typography. While styles can differ a lot based on the prompt, Bubble AI may build your app in its own "fingerprint" style, while apps built through the MCP server take on the style of whichever agent and model you use.
That opens up more variety. If one agent's typical look doesn't match what you have in mind, another may come closer, and you can refine every detail in the visual editor afterward.
Bring in context from your other tools
Your AI agent may also be connected to other MCP servers, like your documents, design files, project notes, or external design resources like Mobbin. Because it can read from those at the same time, it can use that context while it builds in Bubble. For example, it could draft a page based on a spec you've written somewhere else.
Keep full control in the visual editor
Whatever the AI tool builds, it builds with Bubble's own elements, workflows, and data types, so there's no generated code to read. You can open the app in the editor, see exactly how it works, and change anything yourself, whether the work started in Bubble or in another tool.
Usage runs through your AI provider
The MCP server doesn't use Bubble AI, so the AI work doesn't count toward your Bubble AI credits usage. It counts toward your account with the AI provider you're using instead, such as your Claude plan.
How the MCP server compares to Bubble AI
Where you work
In the Bubble editor
In your own AI tool, like Claude, Claude Code, or Codex
AI model
Bubble AI
The model you choose in your AI tool
AI usage
Counts toward your Bubble AI usage
Counts toward your account with your AI provider
What it builds
A Bubble app you can edit visually
A Bubble app you can edit visually
You can use both: Bubble AI while you work in the editor, and the MCP server when you want to work from another agent or try a different model.
What you can do with the MCP server
Once it's connected, your third-party AI agent can work on most parts of your app. Common uses include:
Create a new app
Describe the app you want, and the AI agent builds it in your Bubble account.
Edit an existing app
Add pages, elements, workflows, and data types, or change what's already there.
Debug
Ask the AI agent to find out why something isn't working, and fix it. The AI agent can also check your app's server logs, to see what happened when a workflow ran and find the cause of errors.
Refactor
Ask it to reorganize or simplify parts of your app without changing what they do.
Add sample data
Fill your database with realistic test records, so you can see how your app looks and behaves with data in it.
Set up automations and connect to other MCP services
Create backend workflows that run on a schedule or when data changes, set up API calls to external services, and use information from other MCP servers you've connected, like a spec in your documents.
Complete list of tools
Your third-party AI agent chooses which tools to use based on your prompt, so you don't need to call them yourself. The tables below list everything Bubble's MCP server can do, grouped by category.
list_apps, create_app, check_app_name, whoami
get_app_outline, get_node, get_pages, list_types
create_page, create_element, move_element, set_element_properties
create_style, assign_style, create_color_token
create_event, create_action, reorder_actions
validate_expression, create_global_expression
create_data_type, create_field, create_privacy_rule
create_thing, search_things, update_thing, delete_thing
create_option_set, create_option
create_api_collection, create_api_call
search_plugins, install_plugin
get_preview_link, check_app_issues, create_test_cases
apply_changes, undo, create_savepoint, restore_savepoint
Tool details
Current limitations
A few things aren't available through the MCP server:
Merging branches (but you can specify which branch you want the agent to work on)
Deploying your app
Setting up OAuth2
Buying paid plugins (these have to be bought before the AI tool can install them)
You can still do all of these yourself in the Bubble editor.
Permissions and plans
The MCP server acts as your logged-in Bubble account. When your AI tool makes changes, it uses your Bubble account, so it can only do what you're allowed to do.
Two things follow from this:
Your permissions apply. If you're a collaborator on an app, the AI tool has the same access you do, and no more. Each collaborator who connects their own AI tool works within their own permissions.
Your plan applies. Features that require a certain plan still require it. For example, if your app's plan doesn't include backend workflows, the AI tool can't create them either.
Access to your app's data
When you connect your AI tool, you're asked to approve what it can do with your app's data. The AI tool can read and make changes to records in your development database, and can read the live database.
Development
Yes
Yes
Live
Yes
No
This means the AI tool can add sample data, update records, and help you debug in development, but it can't change the data your users rely on in your live app.
Working with the MCP server
Once you're connected, you work with Bubble the same way you'd work with anything else in your AI tool: describe what you want in plain language. The AI tool decides which of the MCP server's tools to use, makes the changes in your app, and returns a preview link so you can see the result.
Example prompts
Here are a few prompts to get you started:
If you have more than one app, name the app you mean in your prompt, so the AI tool works on the right one.
Reviewing changes
When the AI tool finishes, open the preview link or your app in the Bubble editor to review what it built. Since everything is visual, you can check each element, workflow, and data type, and adjust anything by hand or with another prompt.
You can also look at the change log to get an overview of the changes made.
Undoing changes
Changes the AI tool makes don't go into the Bubble editor's undo history, so pressing undo in the editor won't reverse them. You have two ways to undo:
Ask the AI tool
Ask it to undo, in plain language.
Restore a savepoint
Ask the AI tool to create a savepoint before a big change. If you don't like the result, ask it to restore that savepoint. This replaces every change made since, including any you made yourself in the editor.
A single prompt can result in several changes, so if you want to roll back everything from one request, it's often easiest to ask for that directly. For example:
FAQ: Bubble MCP server
Last updated
Was this helpful?