Introduction
HTML Anything is an open-source editor that turns raw content into finished HTML pages. You give it Markdown, a spreadsheet, a JSON dump or plain notes. It hands that content to a coding agent already installed on your machine, such as Claude Code or Codex, along with a template. The agent writes the page and you watch it render in real time.
It comes from the team behind OpenDesign, it is licensed under Apache-2.0, and the GitHub repo shows about 9,000 stars and 872 forks at the time of writing. The README calls the project "early but real", and that is a fair description. The core loop works as described in the docs. The browser extension, version history and skill marketplace are still on the roadmap.
This review covers what the vendor's homepage and README say, what the pricing actually costs you, and where the limits are. scored.tools has not run it, so claims about how it behaves are attributed to the project's own pages.
Key Features
Many input formats, parsed locally
Per the homepage, HTML Anything accepts Markdown, CSV, TSV, JSON, SQL, Excel and plain text, and detects the format automatically. Tabular data is parsed in the browser with the papaparse and xlsx libraries. The site says "nothing leaves your laptop" for that parsing step. That claim covers parsing only. Once the content goes to your agent, it travels wherever that agent sends its model calls, which for Claude Code, Codex or Gemini CLI means the vendor's cloud.
Uses the agent CLI you already have
The tool looks for supported coding-agent CLIs on your PATH and reuses their existing login, so you don't need a new API key or account. The homepage lists eight: Claude Code, Cursor Agent, OpenAI Codex, Gemini CLI, GitHub Copilot CLI, OpenCode, Qwen Coder and Aider. The README's detailed list adds IBM Bob as a ninth, while its summary still says eight. The difference is small, but it shows the docs are still catching up with the code.
75 skills as plain SKILL.md folders
Templates are called skills. Each one is a folder containing a SKILL.md file, and the project ships 75 of them, grouped by mode (prototype, deck, frame, social, office, doc) and by scenario. The homepage groups the output into nine surfaces: magazine article, keynote deck, résumé, poster, Xiaohongshu card, tweet card, web prototype, data report and Hyperframes video.
Because a skill is a Markdown file, you can open it, read the instructions the agent gets, and fork it. No proprietary template format stands between you and the prompt. Bundled third-party skills keep their own licenses, per the README, so check those before shipping a fork commercially.
Live streaming preview
The agent's stdout streams over server-sent events into a sandboxed iframe, so the layout builds on screen as the agent writes it. The README says plainly that the sandbox allows scripts and same-origin access, which makes the isolation partial. It also says the /api/convert route spawns the CLI with permissive flags and is meant for a single local operator. Keep it on localhost and don't expose it to a network.
Export presets
Export targets per the site:
- WeChat MP: HTML with inlined CSS, ready to paste into the editor
- X, Weibo and Xiaohongshu: a 2x PNG copied to the clipboard
- Zhihu: math rendered as image placeholders, since live LaTeX isn't supported there
- Standalone .html or .png download
Four of those presets target Chinese platforms. Few Western tools cover WeChat's paste quirks or Xiaohongshu's card format. If you publish mainly to a blog, Substack or LinkedIn, you'll mostly use the plain .html and PNG exports.
Setup
No hosted version exists for the full loop. The quickstart is a git clone, pnpm install, then pnpm -F @html-anything/next dev and a browser at localhost:3000. You can deploy the Next.js shell to Vercel, but the agent still runs on your own machine. Anyone comfortable with a terminal will be fine. Anyone who isn't will stall at step one.
Pricing Breakdown
| Plan | Price | What you get |
|---|---|---|
| Open source | Free | Apache-2.0 license, 75 SKILL.md templates, reuses your authenticated coding-agent CLI, no new API key or account |
The software costs nothing. The real cost is the agent behind it. Every page you generate uses tokens or quota from whichever CLI you plug in, billed under that vendor's plan. If you already pay for Claude Code or Codex for development work, HTML Anything adds no new line item. If you don't, you need one of those subscriptions or API accounts first, and the free tier of HTML Anything doesn't change that.
Free CLI options exist on the supported list. Gemini CLI and Qwen Coder have free usage tiers from their vendors, and Aider and OpenCode can point at whatever model provider you configure. Output quality will vary with the model you choose.
Scores
Overall rating: 6.6 / 10.
| Criterion | Score | Basis |
|---|---|---|
| Output quality | 6 | Active project with about 9,000 stars, but template quality is vendor-described only, with no benchmarks or independent reviews |
| Ease of use | 7 | Local-first, no API key, auto-detects agent CLIs; needs an existing agent CLI and lacks a standalone docs page |
| Pricing value | 9 | Free under Apache-2.0; your existing agent subscription is the only dependency |
| API and integration | 5 | Wide input formats and one-click exports, but no documented API or SDK |
| Problem fit | 6 | Does what it claims per the vendor, with adoption to back it, but no independent verification of results |
Pros and Cons
Pros
- Free and Apache-2.0 licensed, so you can fork and modify templates and the app itself
- No separate API key or subscription if you already run a supported agent CLI
- Skills are plain SKILL.md folders that are easy to read, audit and customize
- Export presets cover WeChat, Zhihu, Weibo and Xiaohongshu, which most tools skip
- Tabular parsing happens in the browser, per the site
Cons
- Requires an installed, authenticated coding-agent CLI, and model costs come from that agent's plan
- Output quality depends on the agent and skill you pick, and no independent tests exist yet
- Export presets lean toward Chinese platforms, which matters less if you publish elsewhere
- Partial sandbox isolation and a permissive local API route, both flagged in the README
- Young project with docs that disagree with themselves on details like the CLI count
- No documented API or SDK for scripting it into a pipeline
Who Is It For
The clearest fit is a developer or technical writer who already pays for Claude Code, Codex or Cursor and keeps producing formatted one-off pages: a report from a CSV, a slide deck from meeting notes, a résumé, a social card. HTML Anything gives that agent a set of tested prompts and a live preview, and charges nothing extra.
It also suits anyone publishing to WeChat official accounts or Xiaohongshu. Inlined CSS for WeChat paste and 2x PNG cards are tedious to produce by hand, and here they are presets.
It fits poorly for non-technical users who want a hosted app. Tools like Gamma or Presenton give you a browser UI and an account, with no terminal involved. Teams who want one shared workspace with history and permissions will also find nothing here yet, since version history is still on the roadmap. If you want the wider OpenDesign toolset rather than this single converter, look at Open Design.
Verdict
HTML Anything is a sensible, low-risk install for anyone already paying for a coding agent. The software is free, the license is permissive, and the skill-folder design means you can see and change every instruction the agent receives. Its 6.6 rating reflects a strong value score held back by an unproven output record, thin docs and no API.
Recommendation: if Claude Code, Codex, Gemini CLI or another supported CLI is already on your PATH, clone HTML Anything and try it on a real document. You risk an afternoon and nothing else. If you don't use a coding agent, or you want a hosted product your whole team can log into, skip it for now and come back once the roadmap items ship.
