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
Recommended Servers
playwright-mcp
A Model Context Protocol server that enables LLMs to interact with web pages through structured accessibility snapshots without requiring vision models or screenshots.
Magic Component Platform (MCP)
An AI-powered tool that generates modern UI components from natural language descriptions, integrating with popular IDEs to streamline UI development workflow.
Audiense Insights MCP Server
Enables interaction with Audiense Insights accounts via the Model Context Protocol, facilitating the extraction and analysis of marketing insights and audience data including demographics, behavior, and influencer engagement.
VeyraX MCP
Single MCP tool to connect all your favorite tools: Gmail, Calendar and 40 more.
graphlit-mcp-server
The Model Context Protocol (MCP) Server enables integration between MCP clients and the Graphlit service. Ingest anything from Slack to Gmail to podcast feeds, in addition to web crawling, into a Graphlit project - and then retrieve relevant contents from the MCP client.
Kagi MCP Server
An MCP server that integrates Kagi search capabilities with Claude AI, enabling Claude to perform real-time web searches when answering questions that require up-to-date information.
E2B
Using MCP to run code via e2b.
Neon Database
MCP server for interacting with Neon Management API and databases
Exa Search
A Model Context Protocol (MCP) server lets AI assistants like Claude use the Exa AI Search API for web searches. This setup allows AI models to get real-time web information in a safe and controlled way.
Qdrant Server
This repository is an example of how to create a MCP server for Qdrant, a vector search engine.