Huxly›Docs›Integrations›AI Connector

AI Connector

Use Claude Code, Cursor, Windsurf, or any MCP-compatible AI tool to build inside Huxly — reading files, writing code, and deploying to a live preview. All inference runs on your own AI subscription. Zero Huxly credits consumed.

How it works

Huxly exposes its project tools over the Model Context Protocol (MCP) — an open standard supported by every major AI coding tool. Your AI connects to Huxly's MCP endpoint, authenticates with your API key, and gains access to a set of tools it can call: reading files, writing code, starting the sandbox, checking build logs, and more.

Recommended

All AI inference happens on your own subscription. Huxly only runs the sandbox infrastructure — file storage, build pipeline, live preview. That's covered by your Huxly plan.

Supported tools

Claude.aiMCP

Pro or Max

Claude Code CLIMCP

Free

CursorMCP

Free tier available

WindsurfMCP

Free tier available

AntigravityMCP

Google AI editor

Cline (VS Code)MCP

Free

Continue.devMCP

Free

Zed editorMCP

Free

Warp terminalMCP

Free

Codex CLIMCP

Free

ChatGPTMCP

Plus or higher

Get your API key

Every setup below requires a Huxly API key. Generate one in Settings:

Open Huxly → click your avatar → Settings
Go to the Claude MCP tab
Click Generate API Key
Copy the key — it's shown once only

Heads-up

The AI Connector requires a Developer, Business, or Max plan. Free accounts cannot generate connector keys.

Setup: Claude.ai

Requires Claude Pro or Max subscription.

Open claude.ai → click your name → Settings
Go to Integrations → Add custom integration
Set the name to Huxly
Paste the MCP URL:

https://huxly.app/api/mcp?key=YOUR_API_KEY

Claude can now see and build your Huxly projects inside any conversation.

Setup: Claude Code CLI

Free. Works in any terminal. Add Huxly to your global MCP config:

claude mcp add huxly --transport http https://huxly.app/api/mcp \
  --header "Authorization: Bearer YOUR_API_KEY"

Or add it manually to ~/.claude/claude.json:

{
  "mcpServers": {
    "huxly": {
      "type": "http",
      "url": "https://huxly.app/api/mcp",
      "headers": {
        "Authorization": "Bearer YOUR_API_KEY"
      }
    }
  }
}

Setup: Cursor

Add to your project's .cursor/mcp.json (or global Cursor MCP settings):

{
  "mcpServers": {
    "huxly": {
      "type": "http",
      "url": "https://huxly.app/api/mcp",
      "headers": {
        "Authorization": "Bearer YOUR_API_KEY"
      }
    }
  }
}

Restart Cursor. The Huxly tools appear in the MCP tools panel.

Setup: Windsurf

Add to ~/.codeium/windsurf/mcp_config.json:

{
  "mcpServers": {
    "huxly": {
      "serverUrl": "https://huxly.app/api/mcp",
      "headers": {
        "Authorization": "Bearer YOUR_API_KEY"
      }
    }
  }
}

Restart Windsurf. Huxly tools are available in Cascade.

Setup: Antigravity

Google's AI editor. Add Huxly to ~/.gemini/config/mcp_config.json:

{
  "mcpServers": {
    "huxly": {
      "serverUrl": "https://huxly.app/api/mcp",
      "headers": {
        "Authorization": "Bearer YOUR_API_KEY"
      }
    }
  }
}

Or add it through the editor UI:

Open the MCP Store via the ... dropdown in the side panel
Click Manage MCP Servers → View raw config
Paste the config above, replacing YOUR_API_KEY with your key from Huxly Settings
Save — Huxly tools appear in the editor automatically

Setup: ChatGPT

Requires ChatGPT Plus or higher. Custom plugins sit behind Developer mode, and ChatGPT uses OpenAI's stricter Streamable HTTP transport — so it connects to the dedicated /api/mcp/stream endpoint, not /api/mcp.

Open ChatGPT → Settings → Security and turn on Developer mode
Go to the Plugins tab → click + to open New Plugin
Set the name to Huxly, leave Connection on Server URL, and paste:

https://huxly.app/api/mcp/stream?key=YOUR_API_KEY

Set Authentication to No Auth — the key in the URL is what authenticates you
Tick I understand and want to continue → Create

Heads-up

ChatGPT's plugin dialog can't send a custom Authorization header, so the key travels in the URL instead. It grants exactly the same access as a bearer token — treat the whole URL as a secret.

Available tools

These tools are available to any connected AI client.

list_projectsList all your Huxly projects.

create_projectCreate a new Expo, Flutter, or Swift project.

list_filesGet all file paths in a project.

read_fileRead the content of a specific file.

write_fileCreate or fully overwrite a file.

write_filesWrite multiple files in one call.

edit_fileSurgical string replacement inside a file.

delete_fileDelete a file from the project.

search_filesGrep across all files by string or regex.

get_conventionsGet framework coding rules for the project.

start_sandboxStart the live preview sandbox.

update_sandboxPush all files and wait for rebuild. Returns full build log.

get_build_logsRead the current build log.

get_runtime_logsRead runtime logs (crashes, console output).

get_preview_screenshotTake a headless screenshot of the live preview.

get_figma_designFetch a Figma frame and format it for AI.

exec_in_sandboxRun a shell command inside the sandbox.

get_project_urlGet the Huxly dashboard link for a project.

FAQ

Does using the connector consume Huxly credits?

No. AI inference runs on your own subscription. Huxly only runs the sandbox, which is covered by your plan.

Can multiple AI tools connect at the same time?

Yes. Each connection uses the same API key. You can have Claude Code and Cursor both connected to the same project simultaneously.

Is there a rate limit?

The MCP endpoint has the same rate limits as the Huxly API. For most users this is never hit in practice.

What happens if the sandbox isn't running?

Tools like update_sandbox will return an error telling the AI to call start_sandbox first. The AI will handle this automatically if you tell it to start the preview.

Can I revoke the API key?

Yes — go to Settings → Claude MCP tab → Revoke. The key stops working immediately. Generate a new one and update your AI tool config.

Still stuck?

Open a support thread or check the AI Connector launch post for a full walkthrough.