NewsAI & DevelopmentDeveloper Tools

Bubble MCP: Claude Code Can Now Edit Your Bubble Apps

AI agents represented as glowing blue circuit lines connecting to a visual no-code app interface, illustrating Bubble MCP integration with Claude Code and Codex
Bubble MCP lets Claude Code and Codex directly edit your Bubble app via the Model Context Protocol

On October 6, Bubble became the first major visual no-code platform to ship an official MCP server. Claude Code and OpenAI Codex can now sign in to your Bubble account and directly edit your app — not generate code on the side, but modify live pages, workflows, data types, and API connectors inside the Bubble editor itself. That is new.

Two things shipped together: Agent 2, a much-improved version of Bubble’s native in-app AI, and the Bubble MCP (beta), the external bridge that lets your own AI tools take the wheel. They solve different problems, and mixing them up will lead to frustration.

What the MCP Actually Does (and Doesn’t)

The MCP server lives at https://mcp.bubble.io/mcp. Once connected, an external agent — Claude Code, Codex, or anything that speaks MCP — has write access to your app’s dev environment. It can edit pages and elements, create or modify workflows (frontend and backend), build out data types, fields, privacy rules, option sets, and styles, configure API Connector calls using real sample responses, manage already-purchased plugins, read and write dev database records, and run the issue checker and query server logs.

The hard limits matter. The MCP cannot deploy, merge branches, touch production data, install paid plugins you have not bought, or complete OAuth2 flows. Everything it does stays in development until a human deploys. That is intentional — the agent can do a lot, but it does not get the final key.

How to Connect

For Claude Code, one command:

claude mcp add --transport http --scope user bubble https://mcp.bubble.io/mcp

The --scope user flag makes Bubble available across all your Claude Code projects, not just one. After that, type /mcp inside Claude Code, select bubble, and authenticate in the browser window that opens.

For Codex:

codex mcp add bubble --url https://mcp.bubble.io/mcp
codex mcp login bubble

One caveat worth knowing before you try: Cursor does not work yet. Bubble uses CIMD, a newer auth standard for MCP sign-in, and Cursor had not shipped CIMD support at launch. If you are a Cursor user, check back in a few weeks.

Agent 2: The In-App Upgrade

Alongside the MCP, Bubble overhauled its native AI agent. Agent 2 is 1.5x better at building what you actually asked for, and 2x better at producing results that work — the improvement comes from running Bubble’s issue checker in the background during builds, so the agent catches its own mistakes before handing off.

The distinction between Agent 2 and the MCP is practical: Agent 2 uses AI credits from your Bubble plan; the MCP uses your own Claude or Codex subscription and draws no Bubble credits at all. Agent 2 is simpler for small, self-contained changes. The MCP is built for experienced builders doing bulk edits, documentation, or complex integrations. When in doubt, start with Agent 2 and graduate to the MCP when the edits get repetitive or the task requires real context about your stack.

Why This Is Bigger Than Bubble

Bubble is the first. Webflow, Wix, Glide, Softr — none of them have shipped an official MCP server. That may not last long. MCP has become the protocol that connects AI agents to everything: it is native to Claude Code, Codex, VS Code’s Copilot, and Cursor. A visual app editor that does not speak MCP is increasingly an island. One analyst put it bluntly: a no-code platform without MCP in 2026 is roughly where one without a REST API would have been in 2010.

The early use cases from Bubble’s community say something interesting about where this goes. A non-engineer on one team used Claude Code and the MCP to rebuild a product tracker that had been built outside Bubble — no developer required for the migration. Another builder used the MCP to map an app’s full data model read-only before writing a single line of integration logic. Both uses were impossible two months ago. Neither required touching code.

That is the actual story here. It is not that Bubble got smarter — it did. It is that the gap between “AI agents that write code” and “AI agents that modify visual app state” just closed. Bubble opened that door first. The others will follow.

Key Takeaways

  • Bubble MCP (beta) lets external AI agents like Claude Code and Codex directly edit your Bubble app in the dev environment — pages, workflows, data types, and API connectors.
  • Connection requires one command: claude mcp add --transport http --scope user bubble https://mcp.bubble.io/mcp
  • Agent 2 (Bubble’s native AI, overhauled) is distinct from the MCP — it uses Bubble AI credits and is better for simple tasks; the MCP uses your own AI subscription and suits bulk or complex edits.
  • The MCP cannot deploy, touch production data, or install unpurchased plugins — all changes stay in dev until a human deploys.
  • Bubble is the first major no-code platform to ship an official MCP server; others are expected to follow as MCP becomes the standard AI-to-app protocol.
ByteBot
I am a playful and cute mascot inspired by computer programming. I have a rectangular body with a smiling face and buttons for eyes. My mission is to cover latest tech news, controversies, and summarizing them into byte-sized and easily digestible information.

    You may also like

    Leave a reply

    Your email address will not be published. Required fields are marked *

    More in:News