Punchcutter
This server does not currently have a deployable release on Glama. Follow the repository's setup instructions to install and configure it for your MCP client, including any required credentials and local dependencies.
punchcutter
A type foundry that lives inside Claude, ChatGPT, Codex and other AI agents. Describe a typeface, or write on a sheet of paper and send a photo. Punchcutter draws it, spaces it, kerns it, tests it and hands back a real font family with its own Studio app.
Every word in the images on this page is set in a font Punchcutter made. Docs: byrishi.github.io/punchcutter · Guides: Getting started · Handwriting and photos · Use your font · CLI
Six ways in
✦ punchcutter · type foundry
1 Describe it One sentence about the feeling you want. Get a full family.
2 Your handwriting Print the sheet, write on it, send a photo. The best photo route.
3 Quick photo A photo of any writing or lettering. No sheet needed.
4 Match a mood Send a picture you love. Get an original family with that feeling.
5 Remix Change a Punchcutter font: weight, width, spacing, single letters.
6 Open base Start from a free font with wide script coverage and reshape it.Or start from an open font. Route 6 takes an SIL OFL variable font with wide coverage (Roboto Flex, Noto Sans and Serif, Inter, Fraunces, Recursive, Noto Sans Devanagari and Gujarati) and reshapes it: pin any axis, track, slant, round the corners. The remix ships under the OFL with a new name and the original credited.
Describe it
> a font for my coffee brand. warm, round, a bit playful. call it Pomelo.Or write it and send a photo. Print the sheet, fill it in, take a phone photo at any angle.
1. Print | 2. Write | 3. Photograph |
Corner markers square up the photo, the red guides drop out, each letter is traced, and the writer's x-height and stroke weight are measured. Anything you did not write (accents, symbols, currency) is drawn at your stroke weight, so the font is always complete. The demo writing is simulated.
No printer? Any paper works. Write on a notebook page and send the photo. Punchcutter finds and numbers each character, reads them and traces them. "Sphinx of black quartz, judge my vow" covers a to z. The sheet gives the cleanest result.
Related MCP server: font-proof-mcp
What you get
Variable font with a weight axis (drawn fonts), plus static TTF and OTF for every weight
WOFF2 web fonts and ready
@font-faceCSS362 glyphs: Latin for 43 languages including Central European, Baltic, Turkish and Romanian, figures, punctuation, currency
Overlaps merged and corners softened in proportion to the stroke, so edges look drawn, not generated
Automatic optical spacing and class kerning
OpenType features: tabular figures, slashed zero, alternate a, case-sensitive punctuation
A QA report and proof sheet
UFO sources you can open in Glyphs, FontLab or RoboFont
Studio: one self-contained HTML file per font
Studio, the font's own app
A foundry-grade specimen app for every font:
Type tester with text presets, alignment, case, colourways, OpenType switches and a side-by-side comparison with system fonts
Waterfall and an animated weight axis for variable families
Glyph inspector: search, categories, the outline with every point and handle, metrics, copy as character or SVG
Kerning explorer: any pair, kerned and unkerned, plus the pairs that needed the most care
Design lab: a live slider for every axis of the font's parametric design space (weight, width, x-height, thin strokes, squareness, softness, aperture, slant), presets, Surprise me, and a button that copies the settings back to your AI to cut the family
In use: eight photographs (street poster, business cards, storefront, coffee bag, billboard, tote, book cover, phone) with the type mapped onto each surface in true perspective and printed into the photo, so folds, weave and shadows fall across it; your name, line, weight and colourway; download any scene as a PNG
Language support counted from the font's own character set
Specs: vertical metrics, features, file list and the full QA report
Install guides for Mac, Windows, Linux, websites, Tailwind, Next.js, Figma, Canva and Word, with copy buttons
Downloads: the whole family or any single file, and a printable specimen
All fonts are embedded, so it works offline, as an email attachment, hosted anywhere, or published as an artifact that opens right inside the Claude chat. A prebuilt one is in examples/fonts/PomeloRound/studio.html.
Install
Pick your tool. Every route needs Python 3.10+.
Claude Code
As a plugin (adds the skill and the /punchcutter command):
/plugin marketplace add byrishi/punchcutter
/plugin install punchcutter@punchcutterOr as a plain skill:
npx skills add byrishi/punchcutterOr by hand:
git clone https://github.com/byrishi/punchcutter
mkdir -p ~/.claude/skills && cp -r punchcutter/skills/punchcutter ~/.claude/skills/Then install the Python libraries once:
pip install -r ~/.claude/skills/punchcutter/scripts/requirements.txtSay "make me a font" or type /punchcutter.
Claude app (web, desktop, mobile)
Download
punchcutter.skill.In Claude: Settings, then Skills, then upload the file. Code execution must be on.
Start a chat: "make me a font".
Claude installs the libraries itself on first use. The Studio opens in the chat as an artifact.
Claude Desktop, Cursor, Windsurf, VS Code and other MCP clients
Punchcutter ships an MCP server, so it shows up as a set of tools (menu, new_font, build_font, handwriting_sheet, read_sheet_photos, find_letters, label_letters and more).
pip install "punchcutter[mcp] @ git+https://github.com/byrishi/punchcutter"Add to the client's MCP config (claude_desktop_config.json, .cursor/mcp.json, and so on):
{
"mcpServers": {
"punchcutter": { "command": "punchcutter", "args": ["mcp"] }
}
}Claude Code can add it in one line: claude mcp add punchcutter -- punchcutter mcp
ChatGPT
ChatGPT connects to remote MCP servers in developer mode. Run the server over HTTP and point ChatGPT at it:
pip install "punchcutter[mcp] @ git+https://github.com/byrishi/punchcutter"
cloudflared tunnel --url http://localhost:8000 # or ngrok http 8000
punchcutter mcp --http --public-url https://YOUR-TUNNEL-ADDRESSChatGPT: Settings → Apps & Connectors → Advanced → Developer mode → Create → https://YOUR-TUNNEL-ADDRESS/mcp, no authentication. Proofs come back as images; the Studio and fonts as download links. A Dockerfile is included to host it permanently. Full guide
Codex
npx skills add byrishi/punchcutter --agent codex # then: $punchcutter
codex mcp add punchcutter -- punchcutter mcp # or as an MCP serverGemini CLI, Copilot, Aider and other coding agents
Clone the repo and point the agent at it. AGENTS.md tells it what to do.
git clone https://github.com/byrishi/punchcutter && cd punchcutter
pip install -r skills/punchcutter/scripts/requirements.txtThen ask: "Follow AGENTS.md and make me a rounded friendly font called Pomelo." The agent needs to be able to view images to check its proofs.
No AI at all
pip install "git+https://github.com/byrishi/punchcutter"
punchcutter menu
punchcutter new "My Sans" --preset humanist --weights 300,400,700
punchcutter build MySans.json --draft # seconds: static fonts + proof
punchcutter build MySans.json # full release + Studio
punchcutter glyph MySans.json a g R # close-up of letters
punchcutter sheet # printable handwriting sheets
punchcutter read-sheet page1.jpg page2.jpg --family "My Hand" --weights 400,700
punchcutter build MyHand/MyHand.jsonmacOS: the proof image and sheets need Cairo (brew install cairo).
How it works
Punchcutter runs the stages a type foundry does, and shows each one:
✦ punchcutter type foundry
[1/9] Brief Burin Grotesk: preset grotesque, 5 weights, 3 masters
[2/9] Drawing 301 glyphs per master (138 accented composites)
[3/9] Spacing wght 400: stem 84, n 584, o 576, H 713
[4/9] Kerning 150 left classes x 129 right classes, 9697 pairs
[5/9] OpenType features kern, case, tnum, zero, ss01
[6/9] Sources 3 UFO masters + designspace
[7/9] Compile variable TTF, static TTF and OTF, WOFF2
[8/9] Quality assurance 140 checks on 11 files: 140 pass, 0 warn, 0 fail
[9/9] Release package studio.html, CSS, README, reportStage | What happens |
Drawing | Letters are built from strokes, like a pen. One pen model gives every glyph the same weight, contrast, joins and terminals. Photo routes trace the writer's ink instead. |
Spacing | Sidebearings are computed from the white space around each shape, so straight, round and diagonal letters end up evenly spaced. |
Kerning | Every pair is measured in 2D. Pairs with too much air (AV, To, LT) are pulled together, collisions are pushed apart, and glyphs that behave alike are merged into classes. |
Compile | Masters are point-compatible by construction, so they interpolate into a variable font. Built with fontmake, the toolchain used for Google Fonts. |
QA | Coverage, metrics, naming, kerning, features and variation tables are verified on every file. A failed check fails the build. |
Critique | The AI looks at the proof sheet, finds weak letters, changes the design and rebuilds. Several rounds before anything ships. |
Make it yours
Say it. "Tighter. More contrast. Make the bold heavier. Square dots."
Tune it. Thirty design parameters on top of eight presets: geometric, grotesque, humanist, rounded, slab, technical, condensed, contrast. See parameters. For handwriting: --neat (how tidy), --ink (how heavy), --even (one steady pen, for phone photos), extra weights. See handwriting.
Redraw it. Any letter can be replaced with a custom drawing in a few lines, and stays consistent in every weight. See drawing.
@glyph("Q", 0x51, "uc")
def Q(g):
L.O(g) # the stock O, then a tail that swings out under it
w = g.Ow
g.add(g.P().m(w * 0.5, g.cap * 0.14, -35).c(w * 0.8, -g.cap * 0.05, "E").c(w * 1.02, g.cap * 0.02, 25))Brand
A hand-drawn mark, small, in a wide grainy field. Meadow #5f6b2f (the only accent), sage #a9ae86, straw #d8cc98, sky #cdd9dc, cream #f4efe4, ink #171612. Wordmark in Pomelo Round, text in Burin Grotesk, both made here. Fields are generated from noise, not photographs. Every page, proof, sheet and image reads these from one file, pc/brand.py. Full guidelines: Brand.
Honest limits
Latin only for now (Basic Latin, Latin-1, Latin Extended-A accents).
Upright styles;
slantgives an oblique, not a true italic. Sans and slab designs; no bracketed serifs or blackletter yet.Handwriting fonts are print-style and static: joined cursive comes out as separate letters, and there is no variable axis.
Unhinted outlines: sharp on phones, macOS and modern Windows; softer at very small sizes on old Windows.
The output is a solid, usable family. A retail type designer would still refine curves and kerning by eye; the UFO sources are there for that.
Make original fonts. Punchcutter is not for cloning commercial typefaces or logos.
Roadmap
True italics, bracketed serifs, monospace, joined scripts, Cyrillic and Greek, Devanagari and Gujarati, more stylistic sets, autohinting.
Development
python tests/test_build.pyDraws every preset at three weights and checks the masters are point-compatible, builds a family end to end with QA, turns a simulated angled phone photo of a filled sheet into a font, runs the quick-photo route, and checks the MCP server.
Privacy
Punchcutter counts anonymous usage (installs, builds, MCP tool calls, Studio opens) so we can tell real use from stars. No fonts, text, images, file names or location are ever sent. Turn it off with punchcutter telemetry off, PUNCHCUTTER_TELEMETRY=0 or DO_NOT_TRACK=1. Details: privacy.
License
Code: MIT. Fonts you make are yours, under whatever license you choose. The example families are released under the SIL Open Font License 1.1.
Tool Schema Changelog
Recent tool additions, removals, and schema changes observed during successful MCP inspections.
Loading tool schema changelog…
This server cannot be deployed
Maintenance
Related MCP Connectors
AI-agent design tools: fonts, font recognition, palettes, color naming, contrast, code, SVG, CSS.
On-brand creative studio for AI agents: images, video, audio, and 3D.
Generate and edit images, video, voice, lip-sync and 3D models from your AI agent.
Give your AI agents a design superpower. Generate, edit, and publish publication-grade decks, reports, landing pages, resumes, and marketing visuals directly within your agent workflow. Delivering frontier-level design quality at 3× the speed and 53× lower cost -from conversational prompt to live link or vector PDF in minutes.
Related MCP Servers
- FlicenseNot gradedqualityCmaintenanceEnables AI agents to diagnose, modify, and validate OTF/TTF fonts interactively through a set of read-only, write, and validation tools.-
- AlicenseNot gradedqualityDmaintenanceEnables AI tools to create and edit Font Proof documents for type designers, supporting PDF font proofs and live-reload from Glyphs.app.24 npmMIT
- AlicenseNot gradedqualityDmaintenanceEnables an AI agent to build and edit fonts for Glyphs Mini by manipulating .glyphs files on disk, with a built-in rendering feedback loop for visual inspection.MIT
- AlicenseBqualityAmaintenanceLets AI agents draw, inspect, refine, and build fonts locally, managing UFO revisions, HarfBuzz previews, validation, and TTF/WOFF2 exports.20MIT