Skip to content
Refcat logo

Refcat MCP server

Which MCP servers are Refcat's own, from the official registry and from asking its own domain, whether they answer, and what they offer. A program checked each of these, nobody copied them from a product page.

read 2026-11-01

← Back to Refcat

Whose server it is
  • Registry nameapp.refcat/refcatverified in the MCP registry

The registry name sits under refcat.app. The registry only lets a publisher use that namespace after proving control of the domain through DNS or HTTP.

When we asked itchecked 2026-11-01
  • Stateanswerschecked 2026-11-01
  • Tools18
  • Server calls itselfrefcat
  • Server version1.0.0
  • Protocol version2025-06-18

On 2026-11-01 the server answered initialize and tools/list without a sign-in. First answer in 848 ms.

Should an agent connect itNot proven either way yet
  • Its tools do Search real app screens by intent, such as onboarding or checkout, but not Export complete user flows into Figma.
  • Checked on 3 of the 7 days it takes to judge reliability.

Once connected

  • Parameters are untyped or undescribed in 15 of 18 tools: check arguments before calling.

Said by rules from the measurements above. The likely first call is a model's pick from the tool list; read its description before relying on it.

Measured dailydaily checks: 3, the first on 2026-11-01
  • INITE MCP scoreafter a weeka score needs 7 daily checks: 3 so far
  • What an agent can do with it1 of 2 use cases
  • Answeredmeasuring3 of the 7 daily checks it takes to print a share
  • Median first answer877 ms
  • Space in an agent's context≈ 3,304 tokensthe definitions of all its tools (18), which an agent carries on every turn
  • Tool definitionsB · 86 of 100
  • Typed results0 of 18 toolsdeclared outputSchema: results come typed, read once
  • Parameters without a description: in 15 of 18

What an agent can do with it

  • can do · Search real app screens by intent, such as onboarding or checkout
  • cannot · Export complete user flows into Figma
  • not sure · Give Claude Code or Cursor real screens as design context

Each of the product's use cases, put to Jev against this server's tool list: can an agent do it with these tools?

Worth a look

None of the five warning signs in these definitions.

A program checks the server once a day: initialize and tools/list, no model. Tokens are the definitions' length at about four characters a token. The grade averages six checks of every tool's name, description, schema and annotations; it says how well the tools are described, not whether they work. The score weighs what was measured and nothing else: answering (25%), doing the product's job (25%), definitions (20%), context (10%), use (10%) and how readily a client connects (10%). A part that was not measured is left out, and a hidden instruction caps the score at 40. The state of vendors' MCP servers: method and numbers →

Tools it offers18 tools · read 2026-11-01
  • list_facets

    The filter vocabulary with real counts (number of flows): platform, flowIntents, uiComponents, pageType and screenPatterns. Call this before filtering search_screens or search_flows so filters are real values, not guesses. Optional…

  • get_screen

    One catalog screen with its image and structure. Returns an image block (look at it for color, typography and spacing) and JSON with app, flow, platform, pageType, screenPatterns, uiComponents, keyTexts, dimensions, capturedAt and…

  • search_screens_by_image

    Find catalog screens that look like your images (base64, e.g. a screenshot of the app you are building). Then call get_screen on the matches to compare. Free plans: up to 2 images and 12 results.

  • search_screens

    Search individual UI screens by description, app name or site ('paywall with annual plan first', 'Spotify', 'linear.app'). Returns metadata per screen; set includeImages to also get the top screens as images (max 4, ~1-2k tokens each)…

  • search_flows

    Search whole flows (ordered screen sequences) by intent, app or description. Then get_flow for the steps.

  • search_emails

    Search real product emails (welcome, receipts, onboarding, re-engagement…) by description or brand. Each result is one email with its desktop and/or mobile image; when hasHtml is true, get_email_html returns the original HTML to adapt…

  • get_email_html

    The original HTML of one email from search_emails (tables, font fallbacks, dark mode included). Adapt it to the user's brand instead of writing email HTML from scratch. Pro, or this week's free flows.

  • search_apps

    Find apps in the catalog by name, website or theme ('nubank', 'linear.app', 'meditation').

  • get_app

    Metadata for one app in the catalog. Accepts an appId or a slug like 'spotify'. get_app_flows lists its flows.

  • get_app_flows

    The flows of one app (onboarding, checkout, settings…) in order, with capture date. Accepts an appId or slug. Big apps have hundreds of flows: page with offset. Then get_flow for the steps of one.

  • similar_apps

    Apps like this one, ranked from the app's own flow themes. Accepts an appId or slug.

  • get_flow

    One flow with every step in order. Compact by default: each step has screenDocId (for get_screen), pageType, patterns, components and keyTexts. include 'elements' adds OCR/hotspot boxes per step (large: tens of thousands of tokens for a…

  • get_flow_history

    Earlier captures of the same flow (how it changed over time): capture date, screen count and up to 3 preview images each.

  • similar_screens

    Screens that look like a given screen. Visual, not keyword — no quota. includeImages returns the top matches as images (max 4).

  • list_categories

    The catalog's category rail for a platform, with the key list_apps filters by. Emails, games, store listings and ads have their own rails.

  • list_apps

    Apps in the catalog, optionally filtered by category key (from list_categories) or platform. Platform `email` lists brands with email galleries.

  • list_projects

    The caller's projects, newest first.

  • get_project

    One project with its screens, positions and connections.

How to connect

Add the server to your MCP client config. Claude Desktop, Cursor and most other clients read this shape:

{
  "mcpServers": {
    "refcat": {
      "type": "http",
      "url": "https://mcp.refcat.app/mcp"
    }
  }
}

Add it

Claude Code, in a terminal

claude mcp add --transport http refcat https://mcp.refcat.app/mcp

Add to CursorAdd to VS Code

claude.ai, ChatGPT and other clients: add https://mcp.refcat.app/mcp as a custom connector in the settings.

Each line is something a program found when it went looking: the registry entry, the server's own answer, a spec that parses. A missing line means we did not find it, not that it does not exist.