Claude Design tokens for law firm brand guidelines is the layer beneath every internal-tool build. Anthropic launched Claude Design on April 17, 2026, and the design-tokens system is what makes the difference between every internal tool looking like the firm versus every tool looking like generic SaaS. Per Anthropic's Claude Design announcement, design tokens are named units of visual brand, colors, typography, spacing, components — that the tool applies automatically when generating output. For a law firm that already has brand guidelines (most do), translating those guidelines into design tokens is a 2-4 hour job that pays back on every future internal-tool build. For a firm without formal guidelines, Claude Design's web capture tool can infer tokens from the firm's existing website. This walks through what design tokens actually contain, how to structure them for a law-firm context, and where to store them. Pricing pulled from the Anthropic pricing page.


Design tokens are the smallest, named units of a visual brand. Instead of writing "#0066FF" in code every time you need the firm's primary blue, you reference "primary-color", a token that maps to #0066FF in one central place. Change the token's value once, and every usage updates automatically.

For a law firm, design tokens cluster into five categories:

- Color. Primary brand color, secondary brand color, accent color, status colors (success, warning, error, info), neutral palette (5-9 grays for backgrounds, borders, text), interactive states (hover, focus, disabled). - Typography. Heading font family, body font family, monospace font family, named sizes (display, h1, h2, h3, body, small, caption), named weights (regular, medium, bold), line heights, letter spacing. - Spacing. A scale of named spacing values used consistently for padding, margins, and gaps. Most firms use a 4-pixel or 8-pixel base unit and a scale like xs/sm/md/lg/xl/2xl/3xl. - Layout. Border radii (subtle for buttons, larger for cards), shadow styles for elevation levels, border widths, container widths (max-width for content, max-width for forms). - Component. Specific patterns the firm wants consistent across every tool, a particular button hover state, a specific card treatment, a custom form-field style.

For an internal tool to feel like part of the firm, these tokens need to be applied consistently. Every dashboard, intake form, and triage UI should use the same primary blue, the same heading font, the same spacing scale. Without tokens, every new tool reinvents these decisions, and the result is a visually fragmented internal tool ecosystem. The Claude Design system integration for law firm brand covers the integration step that connects tokens to specific builds.

Translating an existing brand guideline into tokens

Most law firms have a brand guideline document. Some are detailed PDFs from a recent rebrand; some are one-page summaries from years ago. Either is workable. The translation process:

Color. Map every named color in the brand guideline to a token. "Firm Blue #0066FF" becomes a token named `color-primary` with value `#0066FF`. "Accent Gold #D4A017" becomes `color-accent`. Add status colors (success/warning/error) if the brand guideline doesn't include them, the standard set works fine. Add a neutral gray scale (5-9 grays from near-white to near-black). Total: typically 12-18 named color tokens.

Typography. Map the brand guideline's font choices to tokens. "Inter for headings and body" becomes `font-family-heading: Inter` and `font-family-body: Inter`. Define size tokens (`font-size-display: 48px`, `font-size-h1: 32px`, etc.) based on the brand guideline's hierarchy. If the guideline doesn't specify sizes, use a standard scale: 12/14/16/20/24/32/48 in pixels. Total: typically 8-12 named typography tokens.

Spacing. Define a base unit (4px or 8px is standard) and a scale. `space-xs: 4px`, `space-sm: 8px`, `space-md: 16px`, `space-lg: 24px`, `space-xl: 32px`, `space-2xl: 48px`. Total: typically 6-8 named spacing tokens.

Layout. Pull from the brand guideline if it specifies border radii, shadows, container widths. If not, use sensible defaults: `radius-sm: 4px`, `radius-md: 8px`, `radius-lg: 16px`, `shadow-sm: 0 1px 2px rgba(0,0,0,0.05)`, `shadow-md: 0 4px 6px rgba(0,0,0,0.07)`, `container-content: 720px`, `container-form: 480px`. Total: 6-10 named layout tokens.

Components. Add component-specific tokens for any patterns the firm wants enforced, `button-primary-bg: var(--color-primary)`, `button-primary-text-color: white`, `card-bg: white`, `card-border: 1px solid var(--color-neutral-200)`. Total: typically 8-15 named component tokens.

Full translation: 30-50 named tokens, expressible in a 50-100 line JSON or YAML file. Time to translate from a brand guideline: 2-4 hours.

Inferring tokens from an existing website (no brand guideline)

For firms without a formal brand guideline, Claude Design's web capture tool analyzes a target URL and produces an inferred token set. The process:

Step 1: Pick the target URL. Choose your firm's homepage, About page, or practice-area page, whichever best represents your firm's intended visual brand. If different pages on your site use different visual treatments, capture each separately and reconcile manually.

Step 2: Run web capture. In Claude Design, point the web capture tool at the URL. The tool analyzes the rendered page and produces an initial token set: detected colors with frequency, detected typography, detected spacing patterns.

Step 3: Refine the inferred tokens. The web capture output usually includes 30-50 colors detected on the page; you only want 12-18 of these as tokens. Pick the most-used colors, the ones that match your firm's identity, and discard the rest. Same for typography (pick the most-used font families and sizes) and spacing (pick the consistent values, discard one-off measurements).

Step 4: Name the tokens meaningfully. "primary-blue" is better than "#0066FF." "heading-font" is better than "Inter, sans-serif." Named tokens are easier to maintain over time and make Claude Design's outputs more legible to humans reviewing the code.

Step 5: Validate with a sample build. Generate a representative internal-tool screen using the inferred tokens. Check that the output looks like your firm. Iterate on any tokens that produced surprising results. The first sample build is the calibration build.

Total time for inference path: 4-8 hours. Slower than translating from an existing brand guideline (2-4 hours), but faster than developing a brand guideline from scratch (typically a 4-8 week designer project costing $10-$40K). For firms without formal guidelines, the inference path is the right starting point. The Claude Design system integration for law firm brand covers the full integration of tokens into the build workflow.

Where to store the tokens

Tokens can live in several places, each with tradeoffs:

As a Claude Design skill. Per Anthropic's skills documentation, skills are reusable configurations stored inside Claude. A skill containing your firm's tokens can be invoked from any Claude Design session by name. Pros: easy invocation, no GitHub setup required. Cons: harder to version-control across multiple users, less visible to engineers reviewing builds.

As a JSON or YAML file in firm GitHub. Save the tokens as a config file in your firm's GitHub. Reference the file from each Claude Design prompt, or have Claude Code read the file at build time. Pros: version-controlled, easy to update centrally, visible to engineers. Cons: requires a GitHub workflow your legal-ops team may not have.

As CSS variables in a shared stylesheet. For firms with mature web infrastructure, the tokens can live as CSS variables in a stylesheet that Claude Design imports. Pros: directly usable by any web property, including your firm's website. Cons: requires CSS expertise on the build team.

Hybrid approach. Many firms store the canonical tokens in GitHub, expose them as both a Claude Design skill and a CSS file. Updates happen in GitHub; both consumers pull from the same source. This is the most maintainable pattern for firms building 5+ internal tools or running their public website on the same design system.

For a 25-attorney firm building 2-4 internal tools per year, the Claude Design skill alone is enough, the GitHub overhead doesn't pay off until you're managing more builds or more concurrent contributors. For a 100+-attorney firm with a real legal-tech team, the GitHub approach with versioning, pull requests, and review is the right pattern. The Claude Design skill on GitHub for legal-tech builders covers the GitHub-versioning workflow in detail.

The Bottom Line: The verdict: design tokens are the foundation that makes every Claude Design build look like your firm by default. Translating an existing brand guideline takes 2-4 hours; inferring tokens from an existing website takes 4-8 hours. Either way, the tokens pay back on every future build by eliminating per-tool styling work. Store the tokens as a Claude Design skill for casual builds, in GitHub for team-scale operations, or both. Don't try to build internal tools without tokens — the per-tool styling overhead and visual inconsistency will compound.

AI-Assisted Research. This piece was researched and written with AI assistance, reviewed and edited by Manu Ayala. For deeper takes and the perspective behind the research, follow me on LinkedIn or email me directly.