Frequently asked questions

Ground the agent in real, finished websites before it writes code. AI builders converge on the same shapes because they pattern-match the average of the internet; feed them deeply-analysed references (real tokens, motion, section structure, do/don't rules) and the output starts from craft instead. That is exactly what Palate's MCP does; the essay at palatemcp.com/why-ai-sites-look-generic walks the mechanics.

Yes: Palate is a hosted web design MCP. One line connects it to Claude Code, Cursor, Windsurf or any MCP client, and it serves the analysed craft of 2,169 real websites: design tokens, signature moves, inner-page anatomy and do/don't rules. The free tier needs no card.

Palate is a hosted MCP server that feeds AI website builders (Claude Code, Cursor, Windsurf, v0, Lovable) the taste and craft of 2,169 deeply-analysed real websites. Instead of your agent defaulting to generic layouts and stock patterns, it gets structured design intelligence: facets, inner-page depth, do/don't rules, component prompts, and extracted design tokens.

Mobbin and Refero are visual libraries for human designers. Palate is built for machines. Every reference is structured data your AI agent can reason over: layout facets, typography systems, colour tokens, motion patterns, and explicit do/don't guidance. It is not a screenshot browser; it is a design intelligence layer that plugs directly into your build workflow via MCP.

The free plan gives you 50 deep reference reads a month, full access to the reference library, and all MCP tools. Search is always unlimited. That is enough to survey the library properly and ground key sections, and to feel the difference Palate makes; a full site build pulls several times more deep reads, so finishing one needs Pro (unlimited) or a one-time US$39 Build Pass (14 days, no subscription).

One line in your terminal. For Claude Code, sign in with your browser, no token to copy: claude mcp add --scope user --transport http palate https://mcp.palatemcp.com/api/mcp. Adding the server does not open the browser by itself, so finish in Claude Code: run /mcp, select palate, choose Authenticate, and click Allow. The browser finishes on a localhost page, that is Claude Code on your machine confirming the connection, then palate shows connected in your terminal. For CI or any headless setup, add your token instead and it connects straight away, no sign-in step: claude mcp add --scope user --transport http palate https://mcp.palatemcp.com/api/mcp --header "Authorization: Bearer plt_live_...". The --scope user flag is required so Palate is available in every directory, not just the current project. For the full website-builder skill and build-quality hooks, also install the plugin: /plugin marketplace add jake-jiffi/palate-marketplace then /plugin install palate-website-builder@palate, then run /reload-plugins (or restart Claude Code) so the skill loads. For Cursor or Windsurf, add the equivalent JSON block to your MCP config. Sign up at app.palatemcp.com to get your token.

Palate exposes ten refs_* tools: refs_search (find references by vertical, page type, or style), refs_get (deep read of a single reference), refs_for_business (match by business description), refs_match_brief (match a creative brief), refs_similar (find visually similar sites), refs_insights (corpus-wide patterns), refs_get_tokens (extracted design tokens), refs_get_screenshot (captured screenshots), refs_get_astro_recipe (Astro implementation guidance), and refs_list_verticals (browse the collection).

Yes. Palate is a standard MCP server, so any agent or IDE that speaks the Model Context Protocol can connect (Claude Code, Cursor, Windsurf, and similar). For the full build experience, not just look-ups, the doctrine, the Astro stack and the quality gate are portable too: one command (create-palate) scaffolds a gated Astro project for Cursor, Codex, Gemini or Copilot, with the gate wired in as a git pre-push hook and CI. See the build docs. For hosted builders like v0 and Lovable, feed Palate output into your prompts manually until they add native MCP support.

Each reference carries explicit do/don't rules and component prompts. These tell your agent what to borrow, what to avoid, and how to adapt specific components to your project. It is the difference between showing your agent a screenshot and giving it structured craft guidance it can act on.

Sign in at app.palatemcp.com, go to Billing, and choose Pro. Your existing token keeps working with the higher limits immediately. Pro gives you unlimited requests, design token extraction, priority support, and team sharing.

Your API token is hashed with SHA-256 and a server pepper before storage. The plaintext is shown once when you create it and never stored. All requests run over HTTPS. Palate never sees your code, your repository, or your files: the MCP server is read-only and receives only the arguments your agent puts in a tool call. It does log those tool calls, including the free-text brief your agent sends to the brief-matching tools, so we can source the references the library is missing. See our privacy policy for exactly what we store, how long, and how to have it removed.