Install

hypertype installs two things: the skill (typography guidance every agent reads) and the optional MCP server (live font-feature verification, type-system generation, hierarchy critique). The fastest path sets up both. The per-tool steps further down are the by-hand alternative.

Fastest: one file, or one command

Tell any AI assistant:

“Set up hypertype by following https://raw.githubusercontent.com/sceboucher/hypertype/main/INSTALL.md”

It identifies its own tool, runs the install, and verifies. Or run it yourself:

npx -y @sceboucher/hypertype install

That detects your tools (Claude Code, Claude Desktop, Cursor, VS Code), registers the MCP server idempotently, and installs the skill. Add --print to preview without changing anything, or --client claude-code|claude-desktop|cursor|vscode to target one.

Verify: ask the assistant to call analyze_font on “Source Serif 4”. A correct install reports it ships tabular figures but no small caps and no slashed zero (Google serves a reduced subset). The full runbook, including the chat-only paste path, is INSTALL.md.

Claude Code

Install it as a plugin (version-pinned, updates with the repo):

/plugin marketplace add sceboucher/hypertype
/plugin install hypertype@hypertype

Or put the skill straight into your skills folder:

git clone https://github.com/sceboucher/hypertype
cd hypertype && npm run install:skill   # ~/.claude/skills/hypertype, registers as /hypertype

VS Code (GitHub Copilot)

This is the one genuine one-click install. The button installs hypertype as a Copilot custom-instructions file, which Copilot applies automatically when you work on HTML, CSS, or component files.

It uses VS Code’s built-in vscode:chat-instructions/install handler (1.102+), pointed at dist/hypertype.instructions.md. That file is generated from the same source as the skill, so it carries the same guidance. (This installs the instructions, not slab.js; for the justified-headline engine, inline slab.js per Getting started.)

The MCP server (verify fonts for real)

The skill tells a model to verify a font carries a feature before using it. @sceboucher/hypertype lets it actually check, by reading the OpenType features and variable axes straight from the served font file. It also generates context-fit type systems and critiques typographic hierarchy. It runs locally with no API key, and is the only path that can analyze your installed and Adobe-activated fonts.

The one-command install above wires it into every tool it finds. To add it to a single tool by hand, use a button or the config below:

In Claude Code: claude mcp add hypertype -- npx -y @sceboucher/hypertype. Everywhere else, the command is npx -y @sceboucher/hypertype under your client’s mcpServers config. Ten tools (analyze_font, check_css, recommend_css, design_type_system, critique_hierarchy, and more) are documented in the package README.

Cursor

SKILL.md works in Cursor unchanged. Drop the skill folder into a project’s .cursor/skills/ (or your global skills directory) and reload the workspace:

git clone https://github.com/sceboucher/hypertype
mkdir -p .cursor/skills
cp -r hypertype/skill .cursor/skills/hypertype

Cursor loads it when the task calls for typography, the same way Claude Code does.

Codex, Gemini CLI, and other agent CLIs

SKILL.md is the same open standard, so a skill built for Claude Code works elsewhere without changes. Put the skill/ folder in whatever skills directory your tool reads (check its docs for the path), and it will pick the skill up on relevant tasks.

ChatGPT, Claude Desktop, and other chat tools

These don’t install a skill from a repository. Instead, paste the condensed instruction block into your Custom Instructions, a Project, or a Custom GPT. It’s about 890 tokens and carries the same guidance as the full skill:

• skill/paste-block.md

The skill and the paste-block are generated from one source, so they never drift apart.