HorizonX MCP: power up your AI agent with HorizonX components
New · MCP server

Power up your AI agent with HorizonX MCP

Connect your coding agent once. Then describe what you need, like a carousel, a scroll animation or a shader background. Your agent finds the right HorizonX component, customizes it and adds the real code to your project as HTML, Svelte or React.

Set up in under a minute

The HorizonX MCP server connects your coding agent to our catalog. Your agent can search every component, read its settings and export working code straight into your project. Add the server, sign in to HorizonX when your agent asks, and you are done. It works with Claude Code, Codex CLI, Cursor, VS Code with Copilot, Windsurf and Gemini CLI, and with any other client that supports remote MCP.

You need

An active HorizonX membership. See plans and pricing.

  1. Choose your AI tool

    We will show you the exact setup for it.

    Your AI tool
  2. Add the server and sign in

    No key to copy. Claude Code opens HorizonX in your browser; sign in with your HorizonX account and approve access.

    Run this in your terminal

    claude mcp add --transport http horizonx https://horizonx.so/api/mcp
    claude mcp login horizonx

    The second command opens HorizonX in your browser. Sign in and approve, and the tools are ready. You can also sign in later from /mcp inside Claude Code. Claude Code MCP docs

  3. Start building

    Ask in your own words. The more you say about what you need and where it goes, the better the result.

    Try this in Claude Code

    Find a before-and-after image comparison slider on HorizonX and add it to the case study page as a Svelte component.

Using Claude or ChatGPT in the browser?

Add HorizonX as a connector and sign in. Whether you can add connectors depends on your Claude or ChatGPT plan.

Claude app

claude.ai and Claude Desktop

Select Add, then sign in to HorizonX and approve access.

ChatGPT

chatgpt.com and the ChatGPT app

Select +, then Add custom MCP server. Paste the address below, choose OAuth, then sign in to HorizonX and approve access.

https://horizonx.so/api/mcp

See it in action

Ask in plain English, in the chat you already use. Here is what your agent does with each request.

  1. 1

    Add a component to a page

    Say what you need and where it goes. Your agent finds the closest match and writes the file into your project.

    > Find an image comparison slider on HorizonX and add it to the case study page in React.

    horizonx · search_components

    query: "image comparison slider"

    horizonx · get_component

    settings · html, svelte, react

    horizonx · export_component

    framework: "react"

    Added src/components/ComparisonSlider.jsx

    One request: search, inspect, export. Done.

  2. 2

    Compare before you choose

    Ask for options first. Searching and browsing never use up your daily allowance.

    > What hero backgrounds does HorizonX have? Show me three before you pick one.

    horizonx · search_components

    query: "hero background"

    Here are three options and what each one does. Which one do you want?

    Nothing is exported until you choose.

  3. 3

    Customize it before export

    Tell it which settings to change. Your agent checks what can be adjusted and exports the component already configured.

    > Export the skewed carousel as Svelte, with rounder corners and autoplay off.

    horizonx · get_component

    reads every setting and its range

    horizonx · export_component

    framework: "svelte", with your settings

    Added src/lib/SkewedCarousel.svelte

    Your settings, already applied.

  4. 4

    Upgrade an existing page

    Point it at a section you already have. It swaps in components where there is a good match and leaves the rest of your code alone.

    > Rebuild this landing section with HorizonX components wherever one fits.

    horizonx · search_components

    one search for each part of the section

    horizonx · export_component

    only the parts with a good match

    Swapped in two components. Your pricing table is unchanged.

    Your code stays. Components go where they fit.

What your agent can do

Your agent gets six tools: three for components, textures and shaders, and three for the Library, each set listed in the order it uses them. It decides when to call each one. You just describe what you want on the page.

01 · Find

search_components

Describe what you need in plain words, like "cards that tilt on hover" or "ASCII effect on a photo", and your agent gets the components, textures and shaders that match. With no description, it lists everything available.

02 · Inspect

get_component

All the details for one component: what it does, which formats it exports to, and every setting with its current value and allowed range.

03 · Export

export_component

The same code the Export button gives you on the site: standalone HTML, a Svelte component or a React component, with the settings your agent chose already applied.

04 · Find in the Library

search_library

Searches the Library the same way: landing pages and site templates, UI kits, design systems, and image and video assets.

05 · Check the files

get_library_item

What a Library product includes, its format (code, Figma or media) and the files it ships, so your agent knows what it is getting before it downloads. It does not count toward your limits.

06 · Download

download_library_item

Short-lived links to the product files, which your agent fetches straight into your project. Uses one download from your plan, the same as downloading on the site.

Tips

  • Describe what it does, not its name. Search understands what a component does. "Images that stack as you scroll" finds more than guessing at a title.
  • Browse before you export. Ask for options first. Searching and reading details do not count. Only exports and Library downloads count toward your limits.
  • Name the framework. Say HTML, Svelte or React in your request. Otherwise, your agent has to guess from your project.
  • Use one key per machine. Give each laptop or server its own key. That way, revoking one key never disconnects the others.

MCP FAQ

  • Yes. Any active HorizonX membership works. We check your membership on every request, so your agent stops getting results if your membership ends and works again as soon as you renew. There is nothing to set up again.

  • It grows with your plan. Starter: up to 30 different components a day and 100 in any 30 days. Pro: up to 60 different components a day and 200 in any 30 days. Max: up to 150 different components a day and 600 in any 30 days. The day resets at midnight UTC. A component you already exported does not count again, neither that day nor within the 30 days, and searching or reading details never counts. You can see how many you have used in Settings → Developer.

  • Yes. When your agent downloads a Library product, such as a UI kit or a template, it uses your plan’s download limits, the same ones as downloading on the website. It is not part of the component allowance above. Downloading the same product again on the same day does not count again, and searching the Library or checking a product’s files never counts.

  • Any client that supports remote MCP over Streamable HTTP. This page has step-by-step setup for Claude Code, Codex, Cursor, VS Code with Copilot, Windsurf and Gemini CLI, plus the values any other client needs. Most of them connect by signing in to HorizonX in your browser, with no key to copy; Windsurf, and any client that cannot sign in, uses an API key instead.

  • Usually not. Claude Code, Codex, Cursor, VS Code and Gemini CLI can sign in to HorizonX directly: the first time they connect, they open HorizonX in your browser and you approve access. Use a key for clients that cannot sign in, or for scripts and servers with no browser. You can create one in Settings → Developer.

  • Yes. Add HorizonX as a custom connector with the address https://horizonx.so/api/mcp, then sign in to HorizonX and approve access when it asks. There is no key to copy. Whether you can add custom connectors depends on your plan with claude.ai or ChatGPT.

  • Every interactive component on the Components page, every texture look and every shader, which your agent can customize and export as code. A texture comes with its look running on an image: your agent can pass one of yours by its web address, or it uses the sample image. Your agent can also search the Library, which holds UI kits, templates, design systems and assets, and download products from it for you.

  • Yes. We show each key only once and store only a scrambled (hashed) version, so no one at HorizonX can see it. You can have up to five keys and revoke any of them in Settings → Developer. A revoked key stops working immediately.

Included in every plan

Give your agent every component.

Skip building it from scratch. Ask for a carousel, and your agent adds one that already works.