Swatchdog MCP Server
MCP server that checks AI-generated CSS against design tokens, flagging off-brand values and suggesting the closest matching tokens for colors, spacing, radii, and typography.
README
Swatchdog MCP Server
Give your AI agent a visual system to build against — so it writes brand-consistent styles instead of guessing colors, radii, spacing, and type.
✅ Now in the Anthropic Connector Directory. In Claude, open Settings → Connectors, search swatchdog, and click Connect — no URL, no key, no setup.
See it in action
- Live claude.ai run: connector session walkthrough →
- Full test transcript (all green): part-b-transcript.txt →
https://github.com/user-attachments/assets/6ae8b242-d6c7-4a29-8892-27b067835308
What it is
Swatchdog is a Model Context Protocol (MCP) server that runs on-demand design-token drift checks for AI coding assistants — Claude Code, Claude Desktop, Cursor, Google Antigravity, other MCP clients, and claude.ai via the connector.
When your agent asks, Swatchdog checks the generated CSS against a design system — a curated pack or your own tokens — and returns every off-token color, radius, spacing, and type value, each with the closest matching token to use instead. The check is on-demand and the loop is yours: Swatchdog reports drift; it never intercepts writes or changes your code.
Why Swatchdog
AI agents are fast but blind to your design system — they invent off-brand spacing, off-palette colors, and quietly ignore your scales. Swatchdog adds a deterministic drift check to the build loop:
- The agent sets a standard — your own tokens, or a pack.
- It calls Swatchdog on the CSS it generated.
- Swatchdog returns precise fixes (e.g. "use
radius.sm(5px) instead of 7px"). - The agent applies them. The loop, and the review, stay yours.
How is it different from a linter like ESLint?
A linter checks whether your code is valid; Swatchdog checks whether your design is on-system. A linter happily passes color: #ff00ff because it's valid CSS — it has no idea magenta isn't in your palette. Swatchdog catches exactly that, even when the code is flawless. Lint catches broken code; Swatchdog catches broken design.
Can't I just give my agent the tokens, or a design.md?
You can, and it'll still drift. Agents approximate and interpolate even with the rules right in front of them — having the rules isn't the same as following them. Swatchdog verifies conformance in the build loop, where the drift actually happens.
Modes
Pack mode — check against a curated Swatchdog family (Workbench, Showcase, Terminal, plus the free Studio sandbox). Zero-config.
BYO mode — check against your own design system. The agent extracts your tokens (from tailwind.config.js, CSS variables, etc.) and passes them as parameters. Content-only and stateless — your code and tokens are never stored. Free to try on the keyless connector lane, or uncapped with a $12 license.
Intelligent suggestions — Swatchdog doesn't just flag drift, it maps each off-token value to the nearest valid token: color (hex/rgb → nearest theme color), radius, spacing, font-size, and font-family.
Coverage note: checks currently cover hex and standard color formats. HSL-channel representation and complex multi-file token resolution are on the Phase 2 roadmap.
Tools
check_design_drift — connector endpoint · BYO-only · keyless-friendly
reference_tokens(object, required) — your design tokens, e.g.{"color":{"primary":"#b06ed0"},"radius":{"md":"6px"}}code(string, required) — the CSS/markup to check- Returns, per violation: axis · found value · expected token + value · location
check_drift — main endpoint · checks CSS against a pack or a custom token set
content(string, required) — the CSS/markup to checkpaletteId(string, optional) — a pack id (e.g.studio-blue-hour); pack modetokens(object, optional) — your own token set; BYO modesource(string, optional) — telemetry tag (pack,css,tailwind)
Pricing — one-time, no subscriptions
| Tier | What you get | Where |
|---|---|---|
| Free — keyless | BYO checks on a shared, rate-capped lane | connector endpoint |
Free — sandbox key swt_sandbox_studio |
pack checks vs the Studio family | main endpoint |
| $12 — drift-check license | BYO checks, your own uncapped key | both endpoints |
| $19 / $49 — pack or bundle | premium families (Workbench · Showcase · Terminal) + a paid key | both endpoints |
On the main endpoint, a free caller attempting a premium or BYO check gets a structured upgrade payload pointing to swatchdog.dev.
Connect
In Claude (easiest): Settings → Connectors → find swatchdog in the directory → Connect. Nothing to paste, no key.
Other MCP clients (Cursor, Claude Code, Claude Desktop) — add one of these to your MCP config:
Connector endpoint — BYO-only, keyless (add a key to remove the rate cap):
{
"mcpServers": {
"swatchdog-check": {
"type": "http",
"url": "https://swatchdog-connector-970396648818.us-central1.run.app/mcp"
}
}
}
Main endpoint — packs + BYO, bearer key (free sandbox key shown):
{
"mcpServers": {
"swatchdog-sandbox": {
"type": "http",
"url": "https://swatchdog-mcp-970396648818.us-central1.run.app/mcp",
"headers": { "Authorization": "Bearer swt_sandbox_studio" }
}
}
}
Prefer to add the connector to claude.ai manually? Settings → Connectors → Add custom connector → paste the connector URL above → leave auth empty.
Privacy
All checks are on-demand and transient. No source code, files, or tokens are ever stored on our servers. We log only minimal usage metadata — a source tag, which pack, and the finding count — never your license key, your code, or your tokens. Full policy: swatchdog.dev/privacy.html.
Created and maintained by swatchdog.dev · Support: hey@swatchdog.dev · A Ziola Project
Recommended Servers
playwright-mcp
A Model Context Protocol server that enables LLMs to interact with web pages through structured accessibility snapshots without requiring vision models or screenshots.
Magic Component Platform (MCP)
An AI-powered tool that generates modern UI components from natural language descriptions, integrating with popular IDEs to streamline UI development workflow.
Audiense Insights MCP Server
Enables interaction with Audiense Insights accounts via the Model Context Protocol, facilitating the extraction and analysis of marketing insights and audience data including demographics, behavior, and influencer engagement.
VeyraX MCP
Single MCP tool to connect all your favorite tools: Gmail, Calendar and 40 more.
graphlit-mcp-server
The Model Context Protocol (MCP) Server enables integration between MCP clients and the Graphlit service. Ingest anything from Slack to Gmail to podcast feeds, in addition to web crawling, into a Graphlit project - and then retrieve relevant contents from the MCP client.
Kagi MCP Server
An MCP server that integrates Kagi search capabilities with Claude AI, enabling Claude to perform real-time web searches when answering questions that require up-to-date information.
Neon Database
MCP server for interacting with Neon Management API and databases
E2B
Using MCP to run code via e2b.
Exa Search
A Model Context Protocol (MCP) server lets AI assistants like Claude use the Exa AI Search API for web searches. This setup allows AI models to get real-time web information in a safe and controlled way.
Qdrant Server
This repository is an example of how to create a MCP server for Qdrant, a vector search engine.