Changelog

What’s new in StyleRef, newest first. Each entry is tagged by area — Product, MCP, API, CLI, Skill, STYLE.md, or ComfyUI.

2026-09

Product — describe an interface style down to its spacing, radius, and layout. The UI & Web section now covers the parts of a digital product that two apps with the same palette still disagree on: layout archetype and content width, density and spacing scale, a system-wide corner radius, card and input treatments, table and chart style, in-app imagery, loading and empty states, and how the page behaves on scroll. Output Format gained Presentation Context, so a style can also say whether the work is shown as raw UI, in a browser frame, or on a device mockup. Extraction picks all of it up from a screenshot or a description.

2026-08

Skill — use a StyleRef inside Figma. A second skill, styleref-figma, turns a style into real file objects: the palette becomes local color variables, the type rules become text styles, and the interface rules become the things they describe, from surface levels and elevation to button shapes, icons, navigation, and focus rings. The full specification stays in context so new frames hold the look, and where a style sets an accessibility level, contrast pairs are checked against it. Point it at a selection instead and it reads the style already there and writes it down as a portable specification. Applying and capturing need no account or connector. See Agent Skill.

API — find your own styles by tag, not just by name. GET /me/styles?query= now ranks your library the way gallery search ranks the gallery: a term matches a style’s name, tags, category, or description. Each style also comes back with its category, tags, and a heroImage thumbnail, so a picker can show your library and the gallery as one list. See Endpoints reference.

ComfyUI — copy a style prompt straight out of the graph. The Apply node has a 📋 Copy prompt button: queue it once and the whole composed prompt is on your clipboard, ready for ChatGPT, Midjourney, or a STYLE.md file. Four of its six formats are made for pasting somewhere else, and now nothing has to be wired up to read them. See Nodes reference.

Product — prove a style with writing or a clip, not just images. The outputs you attach when you share a style can now be written copy — paste what the style produced, up to 1,200 characters — or a short clip up to 15 seconds. They sit alongside your images in the Made with this StyleRef strip at the top of the share page, so a copywriting or motion style can show what it does in its own medium. See Gallery, sharing & remixing.

API — ask any share link for Markdown. Send Accept: text/markdown to a public style’s /share URL and you get the compact Markdown card — around 2,000 tokens instead of a full HTML page — without rewriting the link. The page also names its Markdown twin in a Link response header, so a fetcher can find it without reading the HTML. See llms.txt and raw Markdown.

MCP — connect your account from inside your client. Ask your agent to save, fork, extract, or publish a style and your MCP client prompts you to sign in, right there — approve the scopes and it keeps working. In Claude Code you can also start it yourself with /mcp. Reading public styles still needs no account. See Authentication & scopes.

Skill · MCP — install the StyleRef Agent Skill in one step. Run /plugin marketplace add StyleRef/styleref-skill then /plugin install styleref@styleref in Claude Code, and your agent learns when to reach for a style and how to apply it in each tool — with the StyleRef MCP server connected in the same install. Other agents can copy the skill folder straight from the repository. See Agent Skill.

2026-07

Product — switch a style item off without deleting it. Every style item, and every section, has an on/off switch — hover an item on a computer, or use Include in output when you open it on a phone. Switched-off items stay on your board but are left out of the compiled StyleRef, in every format and everywhere the style is read: your public page, the API, MCP, and ComfyUI. It’s the fast way to try a style with and without a constraint. See Create a style.

Product — jump from your public style page straight into the editor. When you’re viewing your own shared style, an Edit style button takes you to the board. Your sample outputs now also appear above the board while you edit, with a shortcut back to the publish panel to add or reorder them.

API — extract a style from text or a URL, not just an uploaded image. POST /api/v1/extractions now takes a JSON body with any of text, reference, image_url, or document_url — the same sources the MCP extract_style tool accepts — alongside the existing image upload. See the endpoints reference.

Product — put your best sample output first. Sample images you attach while sharing a style can now be arranged: drag one by its handle, or use the arrow keys. The order you set is the order visitors see, and the first sample becomes the style’s cover when it doesn’t have one. See Gallery, sharing & remixing.

MCP · API · CLI — use your styles from any agent or script. The MCP server at https://styleref.io/api/mcp gives agents 14 tools over Streamable HTTP — anonymous for reading, OAuth 2.1 for anything touching your account. The same capabilities are available as a REST API at https://styleref.io/api/v1, and as npx styleref on the command line. See Connect your client, Overview & authentication, and the CLI reference.

STYLE.md — commit a style to your repo. Compile any style to a STYLE.md file your coding agent can read. There’s a format reference, a linter, and a README badge.

ComfyUI — StyleRef nodes (beta). Your styles load as native ComfyUI nodes, so a style stays in the graph instead of being pasted into a prompt box. Five nodes — Load, Apply, Facets, Reference Images, and Login — plus workflow templates to start from. Public gallery styles need no account, and nothing in the plugin spends credits. See the nodes reference and workflow templates.

Product — connectors. How connectors work, the open-source policy behind them, and how to build your own against the public API. See Connectors.

API · MCP — extract a style from an image you hold locally. POST /api/v1/extractions takes an image upload (up to 8 MB) and returns an id to poll; GET /api/v1/extractions/{id} returns the compiled spec, or saves it straight to your library with ?save_as_draft=true. The source image is kept on the style as an inspiration image. See the endpoints reference.

API · MCP — re-read a past extraction instead of paying for it twice. GET /api/v1/me/extractions lists your extraction history, and GET /api/v1/me/extractions/{id} returns the full block set of one you already own, free to re-read and available compiled with ?format=.

API — any published style as Markdown. Append .md to a /share URL and you get the style as Markdown — an identification header followed by the compiled specification, ready to paste. About 1,000 tokens instead of roughly 28,000 for the web page. See llms.txt & markdown endpoints.

Docs — every page is fetchable as Markdown. Append .md to any path on docs.styleref.io for the same content without the navigation and search payload, with an X-Token-Count header so an agent can budget before reading. The site also publishes its own llms.txt . See llms.txt & markdown endpoints.

API · MCP · CLI · ComfyUI — paste a style’s URL anywhere a reference is accepted. The link in your address bar while editing a style works as a ref, alongside the share slug, the /share URL, and the style’s id. See style references.

API · MCP — richer search results. Search returns up to six hex colors per result, so a picker can paint swatches without a call per style. Search accepts ?offset= and returns hasMore / nextOffset for paging, GET /api/v1/me/saved-styles returns the gallery styles you saved, and GET /api/v1/me/styles accepts ?query= so a search box can filter your own library and the gallery at once.

API · MCP · CLI — your own styles serve your last generation. Fetching one of your own styles returns what you last generated on styleref.io, not your live unsaved edits — the same output the share page shows. A style you have never generated answers 409 styleref_not_generated, asking you to generate it once on the web first. Published gallery styles are unaffected; they’re already frozen snapshots.

Product — choose the cover image for your style. A cover set in Edit info is the one your card shows everywhere. You can pick from the images already on the style — your inspiration uploads and the sample outputs you added when sharing — instead of uploading the same picture again.

Product — credits and plans. A failed extraction refunds its credits automatically, so you’re only charged for successful runs. The free monthly top-up accrues for every full interval, and a single failed payment no longer drops you to the Free plan while the charge retries. Pricing, the app, and these docs all read the same live numbers. See plans & limits.

2026-05

Product — start without an account. Open the style board, fill it in, and extract a style from a reference image before signing up. Your work is saved as you go and carries over when you create an account. See Quickstart.

STYLE.md — an open format for style specifications. A style compiles to a plain-text file with YAML frontmatter and a Markdown body: readable by a person, parseable by a machine, and committable next to your code. See What is STYLE.md and the format reference.

Product — extraction reads several references together. Give extraction more than one reference and it weighs how prominent each one is, then resolves them into a single coherent style instead of averaging them into mud. It covers disciplines beyond image generation — UI and product design, architecture, interior, packaging, editorial, 3D, and data visualization. See Extract a style.

Product — blog posts as Markdown. Append .md to any post URL for the plain text, and every post has a Copy for AI button that copies it with a source header attached, ready to paste into a chat.