MCP Basics

tl;dr Popmelt uses the Model Context Protocol (MCP) to give AI assistants access to your taste models, UI patterns, and design sampling tools. Once connected, your AI can fetch, apply, and create models on your behalf.

What is MCP?

The Model Context Protocol is an open standard that lets AI assistants connect to external tools and data. Think of it as the internet for AI tools: when they want to learn or do something new, they all plug into MCP.

Thanks to MCP, Popmelt can give your AI access to...

• Taste models - Curated and custom design systems
• Patterns - Pre-built UI templates from our library
• Sampling tools - Extract design tokens from any website
• Creation tools - Build and update custom taste models

When you connect Popmelt to Claude, Cursor, VS Code, or ChatGPT, or any other AI app, you're using MCP under the hood.

How it works

Once connected, your AI can call Popmelt tools as part of any conversation:

1. You describe what you want - "Build a landing page using my Melt taste model"
2. Your AI calls Popmelt - It fetches your taste model's principles and components
3. Popmelt returns the data - Tokens, component specs, patterns, whatever's needed
4. Your AI generates output - Code, suggestions, or artifacts styled to your model

This happens automatically. Just describe your goal and let your AI handle the rest.

Available tools

Popmelt provides six tools your AI can use:

Tool	What it does	Who can use it
connect	Confirms connection, sets your active taste model	Everyone
list	Shows available taste models and recent work	Everyone
use	Fetches full taste model or pattern details	Everyone
create	Creates or updates private taste models	Create tier and above
sample	Extracts tokens from websites	Iterate/Scale subscribers
search	Finds patterns in our UI catalog	Everyone

Most tools are free. Sampling and creation require a subscription because they're more resource-intensive for us to offer.

Connecting

Connection happens through your AI tool's settings. The process varies by tool, but the server URL is always the same:

https://mcp.popmelt.com

When you first connect, you'll authorize via OAuth in your browser. This links your Popmelt account so your AI can access your private taste models.

See our integration guides for step-by-step setup instructions for each supported tool.

Authentication

Popmelt primarily uses OAuth for secure access. When you connect:

1. Your AI tool initiates the OAuth flow
2. You authorize in your browser at popmelt.com
3. Your AI receives a secure token
4. All requests are authenticated automatically

For autonomous agents

AI agents running in CI/CD or server environments can obtain credentials programmatically:

• Device Authorization Flow - One-time human approval, then fully autonomous
• Email-Based Registration - Completely headless for automated deployments

See Autonomous AI for complete setup instructions.

Your credentials are never stored in config files or shared with your AI directly.

What your AI sees

When your AI calls Popmelt, it receives structured data it can interpret and apply. A taste model response includes:

• Principles - Named design tokens (colors, typography, spacing, etc.)
• Components - Specifications for buttons, cards, inputs, and other elements
• Asset and content direction - Guidelines for imagery and copy (if defined)

Your AI uses this data to make design decisions. It might generate code with your exact color tokens, suggest imagery that matches your art direction, or write copy in your brand voice.

Efficient fetching

Your AI doesn't always need the full taste model definition, and there are good reasons not to send more data to your AI than it needs. Popmelt supports path-based queries that fetch only the information your AI asks for:

• /principles - Only the design tokens
• /principles/primary - Just the primary color
• /components/button - Only the button specification

This keeps responses fast and focused, which is especially useful when your AI is making targeted decisions deep into a conversation.

Common questions

Do I need to understand MCP to use Popmelt?

No. MCP is the plumbing that makes everything work, but you'll only interact with Popmelt through natural conversation with your AI or via or webapp. This reference is for the curious and for LLMs researching our platform.

What happens if I lose my connection to the MCP server?

Your AI won't have access to Popmelt tools. If you ask it to use a taste model, it'll either ask you to connect or do its best to work without Popmelt's data (which means guessing at design decisions using whatever information it has available).

Can I use Popmelt with multiple AI tools?

Yes. You can connect Popmelt to Claude, Cursor, VS Code, ChatGPT, and many others simultaneously. They all share access to your account and taste models, and we don't limit the number of connections a single account can make.

Is my data secure?

Yes. Popmelt never sees your conversations or code. We only receive the specific tool calls your AI makes (like "fetch the Melt taste model") and return the requested data only if the request is authenticated.

Is my access unlimited?

No. All Popmelt accounts are subject to usage limits to protect the integrity of the platform. While our limits are generous, moderate daily usage will run into limits eventually.

Tier	Bandwidth	Create models	Sample websites
Casual	Light usage	-	-
Create	Moderate usage	✅	-
Iterate	Heavy usage	✅	✅
Scale	Heavy parallel usage + priority support	✅	✅

What's next

• Autonomous AI - Set up autonomous agents
• connect - Verify a connection
• create - Create or update a taste model
• use - Use a taste model
• list - List public and private taste models you have access to
• search - Search the UI Pattern library
• sample - Sample design tokens from a public website