MCP Server / Now Live

The first Design Intelligence Layer your AI agent can natively speak.

Give your AI agent live access to your expression infrastructure. Tokens, brands, and guardrails, directly from Claude Code or Cursor. No manual lookups. No copy-pasting values.

Six tools. Zero friction.

Create Brand

create_brand

Create a new brand from natural language. Express your vision once, get production-ready tokens back.

Get Tokens

get_tokens

Retrieve your full token set. Colors, typography, spacing, radii, shadows. Always current, always resolved.

Push Overrides

push_overrides

Push runtime overrides to any token. Change brand colors, swap typography, adjust spacing on the fly.

Resolve Conflicts

resolve_conflicts

Detect and resolve token conflicts between brands, themes, and component-level overrides.

Accessibility Guardrails

accessibility_check

Validate contrast ratios, touch targets, and WCAG compliance. Violations are caught before they ship.

Style Audit

lint_hardcoded_styles

Scan code for hardcoded values. Find hex colors, pixel values, and magic numbers that should be tokens.

claude-code

Add this MCP server to my claude_desktop_config.json:

"designless": { "command": "npx", "args": [ "mcp-remote", "https://mcp.designless.app/mcp", "--header", "x-api-key: YOUR_API_KEY" ] }

Paste into Claude Code. It handles the rest.

designless MCP vs Google Stitch MCP

Google Stitch ships an MCP server with 5 tools: build_site, get_screen_code, get_screen_image, extract_design_context, and project management. It connects Stitch's canvas to coding agents. The designless MCP server connects expression infrastructure to coding agents. Different sources, different paradigms.

Source

Stitch canvas (designs)

Expression infrastructure (contracts)

Tools

5 tools (build, export, extract)

6 tools (create, resolve, enforce, audit)

Output

Screen code + images + design.md

Live tokens + brand contracts + guardrails

Data direction

One-way (canvas → code)

Bidirectional (brand ↔ agent ↔ code)

Determinism

Non-deterministic

Deterministic

Governance

None — no enforcement

Agentic guardrails + accessibility checks

Requires

Stitch canvas (Google account)

LESS Studio API key

Works with

VS Code, Cursor, Claude Code, Gemini CLI

Claude Code, Cursor, Lovable, Cowork

▸ Full Google Stitch comparison ▸ design.md vs LESS Tokens ▸ Design Agent comparison ▸ How to Train Your Agent

Skills / The Orchestration Layer

MCP tools are atomic operations. Skills chain them into workflows. A skill tells an agent: read the brand, check accessibility, apply tokens, validate output. Skills built on skill.design can orchestrate the designless MCP alongside other servers. Figma MCP for canvas context. LESS MCP for design judgment. One skill, multiple intelligence sources.

▸ Figma Skills on skill.design ▸ How Figma Skills Work ▸ skill.design Registry

Agents Built on Expression Infrastructure

Expression infrastructure is designed to be consumed by autonomous agents at runtime. Agents read tokens, execute skills, and resolve against the same system the MCP server exposes — making design decisions without human intervention. More to come.

▸ Expression Infrastructure ▸ designtoken.md