LinkedIn Formatter

Format LinkedIn posts with Unicode styles from a live preview beside your editor.

The problem

LinkedIn has no bold button. Every "𝐛𝐨𝐥𝐝" post you have ever seen is a Unicode trick: those are different characters, not styled ones. So when an AI CLI drafts your post, you end up prompting it again and again. Make that word bold. No, the other one. Now undo it. A round trip through a language model, for one word.

This extension ends that loop. The post lives in a plain .linkedin file. A faithful LinkedIn card renders beside the editor and updates on every keystroke, whether you typed it or your AI did. Select text in the card, click bold, and the file now contains the real Unicode characters. Copy, paste into LinkedIn, done. What you saw is exactly what ships.

Why a file

Because a file is the one interface humans and AI agents already share. There is no export step and no sync problem: the Unicode styling IS the text. Your AI edits the same file you click around in, and the marketplace listing carries instructions the AI reads so it knows to create .linkedin files on its own.

What it does

• Live preview card that mirrors LinkedIn's feed, palette-matched to your editor theme, with your name pulled from git config
• Full styling from the rendered card itself: select text in the preview, pick one of eleven font families, toggle bold or italic, add strikethrough or underline, insert emoji, clear formatting, place the cursor with a click, undo with the usual keys
• Type directly in the card, like a word processor: styles continue as you type, a family or bold picked at the caret latches for the next run, Enter carries formatting into the new paragraph, arrows and Home/End navigate the wrapped card, Ctrl+Z takes back whole words, and typing over a selection replaces it - every keystroke landing in the file as real Unicode
• 18 Unicode letterform styles plus two combining marks, with toggles that know what Unicode can and cannot do: buttons disable only where no variant exists, and selections survive every edit
• A character counter that counts UTF-16 units, because that is how LinkedIn actually counts, verified against the real composer
• Zero network egress: no telemetry, no analytics, no requests of any kind, enforced by tests and a nonce-locked Content Security Policy

The full user guide (styles table, shortcuts, settings) lives on the marketplace listing.

Install

code --install-extension kaushiknaru.linkedin-formatter

Or search LinkedIn Formatter in the VS Code Extensions view. Cursor, Windsurf and other VS Code derivatives take the same id from Open VSX, with cursor or windsurf in place of code.

Then create any file ending in .linkedin and the preview opens by itself.

For AI agents: the MCP server and CLI

The same conversion core ships as an MCP server and a CLI, so an agent can style a post and open the preview without ever being taught the file convention.

MCP server. Claude Code registers it in one line:

claude mcp add -s user linkedin-formatter -- npx -y linkedin-formatter-mcp

Claude Desktop, Cursor, Windsurf and Codex take the JSON, which fetches the server on demand so there is nothing to install first:

{
  "mcpServers": {
    "linkedin-formatter": {
      "command": "npx",
      "args": ["-y", "linkedin-formatter-mcp"]
    }
  }
}

Restart the client afterwards. MCP servers attach at session start, so one registered mid-session does nothing until the client is restarted.

CLI, for shells, scripts and any agent that is already holding a terminal:

npm install -g linkedin-fmt
linkedin-fmt style "launch day" -f sans-serif -b

Or without installing anything, which also works the moment an MCP server is registered but not yet loaded:

npx -y linkedin-fmt style "launch day" -f sans-serif -b

Both are single-file bundles with zero dependencies and zero network access, published as linkedin-formatter-mcp and linkedin-fmt. mcp/README.md has the per-client setup details.

Under the hood

The conversion core (src/lib/) is pure TypeScript with zero vscode imports: style tables transcribed from the Unicode charts with their exception tables, an 11-family bold/italic matrix, lossless round-trip conversion, and fail-closed mapping (a character with no Unicode equivalent stays unchanged, never a lookalike). It runs headless, which is why the 638-test suite finishes in under a second.

The preview is a hardened webview: strict CSP, per-render nonce, every message validated at the trust boundary, CSS and JS inlined so a render can never appear unstyled.

Contributing

git clone https://github.com/MrTig-afk/linkedin-formatter
cd linkedin-formatter
npm install
npm run test:unit && npm run lint && npm run build

F5 launches the Extension Development Host. Packaging uses the marketplace-specific readme:

npx @vscode/vsce package --readme-path README.marketplace.md

CONTRIBUTING.md has the test gates and the invariants (zero egress, fail-closed conversion, validated webview messages) that pull requests must keep intact.

Connect

License

MIT