# DESIGN.md — Anthropic

> Book-cloth terracotta on ivory paper, serif-led editorial layouts — AI presented with humanist warmth.
> Distilled by design.live · AI · anthropic.com
> Agent prompt: "Use the design.live MCP and make it look like Anthropic."

## Brand context
AI research and products that put safety at the frontier

Anthropic builds frontier AI systems and products, including the Claude family of models, for organizations, researchers, and civic groups that need reliable, interpretable, and steerable AI. Its edge is combining safety-focused frontier research with safety techniques applied directly to products—so teams can deploy models for high-stakes work, as shown by Claude helping NASA’s JPL plan route waypoints for the Perseverance rover on Mars.

Audiences:
- Enterprise and business teams adopting Claude for coding, customer support, and productivity workflows
- AI developers and system builders using Anthropic’s developer products and models to build agentic and coding applications
- Public-sector and institutional organizations (governments, research labs, nonprofits) evaluating and deploying safer AI

## Assets
logo (for light surfaces): https://cdn.design.live/v1/domain/anthropic.com/logo?theme=dark
logo (for dark surfaces):  https://cdn.design.live/v1/domain/anthropic.com/logo?theme=light
icon:                      https://cdn.design.live/v1/domain/anthropic.com/icon

## Palette
#CC785C    accent
#141413    dark
#F0EFEA    brand
#FFFFFF    light

## Type
title    Anthropic Sans (custom)
body     Anthropic Serif (custom)

## Type scale
display:    Styrene + Tiempos serif
weights:    400 · 500 · 600
scale:      58 / 38 / 26 / 18 / 15
tracking:   normal, book-like leading

## Signature moves
- Ivory paper, never pure white
- Serif body for long-form trust
- Hand-drawn illustration accents
- Terracotta used sparingly, warmly

## Evidence coverage (private beta)
screens indexed:       37
screens VLM-annotated: 37
workflows mapped:      3
workflows annotated:   3
platforms:             iOS, Web

Evidence returned by the MCP is deduplicated, ranked for the request, and linked to source and permission provenance. Counts reflect the canonical evidence-store snapshot, not invented adoption or asset totals. Screen imagery is delivered through authenticated, metered MCP calls; it is not distributed on public pages.

## Retrieval
Ask the design.live MCP for the project brief, constraints, and desired Anthropic qualities. It returns ranked evidence and an agent-ready implementation plan; do not copy a source screen wholesale.

— design.live/brands/anthropic