jsx-notation
Compresses React/Next.js files, HTML, and SVG into a compact JSXN notation optimized for AI assistants, reducing token consumption by ~40%. Provides an MCP server with tools to read, encode, and decode JSXN.
README
jsx-notation
Compress React/Next.js files, HTML, and SVG by ~40% for AI assistants. An MCP server and library that converts JSX/TSX, HTML, and SVG into JSXN — a compact notation optimized for LLM token consumption.
Why
Every time an AI assistant reads your React components, HTML, or SVG files, it wastes tokens on closing tags, repeated className/class attributes, and verbose props. That's context window space that could be used for actual reasoning.
JSXN strips the redundancy while keeping the meaning:
// Before: 278 chars
<Modal isOpen={showModal} onClose={handleClose}>
<div className="modal-body">
<h2>{title}</h2>
<Button disabled={!selected} onClick={handleSubmit}>Submit</Button>
</div>
</Modal>
// After: 170 chars
@C Modal=M, Button=B
@P onClick=k, disabled=x, isOpen=io, onClose=oc
M {io:showModal, oc:handleClose}
.modal-body
h2 (title)
B {x:!selected, k:handleSubmit} "Submit"
Works for HTML too:
<!-- Before: 340 chars -->
<table class="data-table">
<thead>
<tr><th>Name</th><th>Email</th><th>Role</th></tr>
</thead>
<tbody>
<tr><td>Alice</td><td>alice@example.com</td><td>Admin</td></tr>
<tr><td>Bob</td><td>bob@example.com</td><td>User</td></tr>
</tbody>
</table>
// After: 188 chars
table.data-table
thead
tr
th "Name"
th "Email"
th "Role"
tbody
tr
td "Alice"
td "alice@example.com"
td "Admin"
tr
td "Bob"
td "bob@example.com"
td "User"
Works for SVG too:
<!-- Before: 306 chars -->
<svg xmlns="http://www.w3.org/2000/svg" viewBox="0 0 24 24" fill="none"
stroke="currentColor" stroke-width="2" stroke-linecap="round"
stroke-linejoin="round">
<circle cx="12" cy="12" r="10"></circle>
<line x1="12" y1="8" x2="12" y2="12"></line>
<line x1="12" y1="16" x2="12.01" y2="16"></line>
</svg>
// After: 251 chars
svg {xmlns:http://www.w3.org/2000/svg, viewBox:0 0 24 24, fill:none, stroke:currentColor, stroke-width:2, stroke-linecap:round, stroke-linejoin:round}
circle {cx:12, cy:12, r:10}
line {x1:12, y1:8, x2:12, y2:12}
line {x1:12, y1:16, x2:12.01, y2:16}
Indentation replaces closing tags. .class and #id work like CSS selectors. Frequent components, props, and CSS classes get short aliases. Values with commas are quoted to avoid delimiter confusion.
Quick start: MCP server
The main use case — let your AI assistant read JSX/HTML/SVG files in compressed form.
Claude Code:
claude mcp add jsx-notation -- npx jsx-notation
That's it. Restart Claude Code and the tools will be available. No
npm installneeded —npxdownloads everything automatically.
VS Code (.vscode/mcp.json):
{
"servers": {
"jsx-notation": {
"command": "npx",
"args": ["jsx-notation"]
}
}
}
Add this file to your project and restart VS Code. It will detect the server on its own — no extra setup needed.
Cursor (~/.cursor/mcp.json or .cursor/mcp.json):
{
"mcpServers": {
"jsx-notation": {
"command": "npx",
"args": ["jsx-notation"]
}
}
}
Just add this to your config file and restart Cursor. It will pick up the server automatically — nothing else to install.
Troubleshooting: If the server shows "No tools or prompts" with a red dot, Cursor can't find
npxin its PATH (common with nvm). Fix it by replacing"npx"with the full path — runwhich npxin your terminal and use that value (e.g."/Users/you/.nvm/versions/node/v22.22.0/bin/npx").
Windsurf (~/.codeium/windsurf/mcp_config.json):
{
"mcpServers": {
"jsx-notation": {
"command": "npx",
"args": ["jsx-notation"]
}
}
}
Same format as Cursor. Add it, restart Windsurf, and it's ready.
Cline, Continue, Amazon Q, and JetBrains IDEs (IntelliJ, WebStorm, etc.) also use the same
mcpServersformat. Paste the JSON above into your MCP settings.
Zed (~/.config/zed/settings.json):
{
"context_servers": {
"jsx-notation": {
"command": "npx",
"args": ["jsx-notation"]
}
}
}
Add this inside your Zed settings file and restart. Zed uses
context_serversinstead ofmcpServers.
The server exposes four tools:
| Tool | Description |
|---|---|
read_jsxn |
Reads a .jsx/.tsx/.js/.ts/.html/.svg file and returns its JSXN encoding |
encode_jsxn |
Encodes raw code as a string (file or snippet mode, jsx or html format) |
decode_jsxn |
Decodes JSXN back to JSX/TSX or HTML (jsx or html format) |
write_jsxn |
Decodes JSXN and writes the result to a file (auto-detects .html/.svg for HTML output) |
Guiding the AI
Add this to your CLAUDE.md or .cursorrules so the assistant prefers JSXN:
When you need to read .jsx, .tsx, .js, .ts, .html, or .svg files for context,
use the read_jsxn tool from the jsx-notation MCP server. It returns JSXN compact
notation (~40% fewer tokens) that you understand perfectly.
Library usage
For programmatic use: npm install jsx-notation — exports encode, encodeFile, encodeHTML, decode, decodeFile.
Notation reference
Elements and attributes
| JSX/HTML | JSXN |
|---|---|
<div className="a b"> / <div class="a b"> |
.a.b (implicit div) |
<span id="x"> |
span#x |
<h2>{title}</h2> |
h2 (title) |
<span>Hello</span> |
span "Hello" |
disabled (boolean prop) |
{disabled} |
{...props} (spread) |
{...props} |
Patterns (JSX only)
| JSX | JSXN |
|---|---|
{cond && <X/>} |
?cond > X |
{a ? <A/> : <B/>} |
?a > A | B |
{items.map(i => <X/>)} |
*items > X |
<>...</> |
_ |
File-level compression (encodeFile)
| Code | JSXN |
|---|---|
import { X } from 'mod' |
@I mod: X |
import type { T } from 'mod' |
@T mod: T |
const [x, setX] = useState(0) |
@state x = 0 |
const ref = useRef(null) |
@ref ref = null |
const x = expr |
x = expr |
| JSX return boundary | --- |
HTML/SVG-specific
| HTML/SVG | JSXN |
|---|---|
<!DOCTYPE html> |
!DOCTYPE html |
<br> / <img> (void elements) |
br / img {src:...} |
class="foo bar" |
.foo.bar |
<circle cx="12" /> (SVG self-closing) |
circle {cx:12} |
font-family="system-ui, sans-serif" |
font-family:"system-ui, sans-serif" |
When decoding with { format: 'html' }:
classis used instead ofclassName- All attribute values are strings (no
{expression}) - Void elements use
<br>(not<br />) - SVG elements without children use
<path ... />(XML self-closing) - Empty non-void elements use
<div></div>(not<div />) - Values with commas are quoted in props:
{key:"val, val2"}
Alias headers
| Header | Purpose | Example |
|---|---|---|
@C |
Component aliases | @C Button=B, Modal=M |
@P |
Prop aliases | @P onClick=k, onChange=g |
@S |
CSS class aliases (Tailwind) | @S items-center=ic, font-medium=fm |
Contributing
Found a bug or have an idea? Open an issue.
Disclaimer
This software is provided "as is", without warranty of any kind. See the MIT License for full terms.
JSXN is an experimental tool that compresses JSX/TSX, HTML, and SVG source code into a compact notation for use with AI/LLM systems. The encoded output is a lossy representation and may not preserve all semantic details of the original code. The author makes no guarantees about the accuracy of the encoding, the behavior of AI systems consuming JSXN output, or fitness for any particular use case.
The MCP server component reads files from your local file system. Only .jsx, .tsx, .js, .ts, .html, and .svg files are accepted, with a 10 MB size limit. You are responsible for ensuring it is configured appropriately for your environment. Use at your own risk.
License
MIT - Sebastián Maciel
推荐服务器
Baidu Map
百度地图核心API现已全面兼容MCP协议,是国内首家兼容MCP协议的地图服务商。
Playwright MCP Server
一个模型上下文协议服务器,它使大型语言模型能够通过结构化的可访问性快照与网页进行交互,而无需视觉模型或屏幕截图。
Magic Component Platform (MCP)
一个由人工智能驱动的工具,可以从自然语言描述生成现代化的用户界面组件,并与流行的集成开发环境(IDE)集成,从而简化用户界面开发流程。
Audiense Insights MCP Server
通过模型上下文协议启用与 Audiense Insights 账户的交互,从而促进营销洞察和受众数据的提取和分析,包括人口统计信息、行为和影响者互动。
VeyraX
一个单一的 MCP 工具,连接你所有喜爱的工具:Gmail、日历以及其他 40 多个工具。
graphlit-mcp-server
模型上下文协议 (MCP) 服务器实现了 MCP 客户端与 Graphlit 服务之间的集成。 除了网络爬取之外,还可以将任何内容(从 Slack 到 Gmail 再到播客订阅源)导入到 Graphlit 项目中,然后从 MCP 客户端检索相关内容。
Kagi MCP Server
一个 MCP 服务器,集成了 Kagi 搜索功能和 Claude AI,使 Claude 能够在回答需要最新信息的问题时执行实时网络搜索。
e2b-mcp-server
使用 MCP 通过 e2b 运行代码。
Neon MCP Server
用于与 Neon 管理 API 和数据库交互的 MCP 服务器
Exa MCP Server
模型上下文协议(MCP)服务器允许像 Claude 这样的 AI 助手使用 Exa AI 搜索 API 进行网络搜索。这种设置允许 AI 模型以安全和受控的方式获取实时的网络信息。