MCP

MCP Overview

The Model Context Protocol server for Gummble — stream design context into any agent.

The Gummble MCP server lets your AI tools call Gummble like a teammate. It implements the Streamable HTTP transport with OAuth 2.1 Resource Server semantics, so any standards-compliant MCP client (Claude Desktop, Cursor, Windsurf, Codex, v0, and other agent tools) can pull real product UI references into its workflow.

https://mcp.gummble.com/mcp

Quick setup

Per-client pages describe supported custom setup routes, not marketplace listings or platform partnerships. Follow the platform-specific guide because labels, trust prompts, token storage, and workspace policies differ.

How auth works

  1. Your client POSTs to https://mcp.gummble.com/mcp without a token.
  2. The Worker replies 401 + WWW-Authenticate: Bearer realm="mcp", resource_metadata="https://mcp.gummble.com/.well-known/oauth-protected-resource/mcp".
  3. The client fetches the resource metadata → finds https://api.gummble.com as the authorization server.
  4. The client runs Dynamic Client Registration → opens an authorize URL in your browser → you sign in on gummble.com → consent → callback.
  5. The client exchanges the code for an access token (audience-bound to https://mcp.gummble.com/mcp) and retries the original MCP request.

Client labels, trust prompts, token storage, reconnect behavior, and rollout policies differ. Follow the relevant client guide instead of assuming every MCP interface exposes the same controls.

If you're building an app that stores tokens and calls Gummble MCP on behalf of your users, use the Build an integration guide.

If install or auth fails, start with Troubleshooting.

Available tools

Browse the full list at Tools. High-level groups:

  • gummble_search_* — apps, screens, website sections, flows, patterns, microcopy, and App Store marketing screenshots
  • gummble_get_* — app, screen, flow, and store-screenshot details
  • gummble_find_similar_screens — visual alternatives from a selected reference

Rate limits

Per-user, tracked on the Worker via Cloudflare KV. See Rate limits.