linkedin-fmt

Style LinkedIn posts with real Unicode from the command line.

LinkedIn's composer has no formatting controls. Every "bold" LinkedIn post is a Unicode substitution: those are different code points, not styled ASCII. Markdown does nothing, because **bold** pastes in as literal asterisks.

This converts text to the characters that actually survive the paste.

$ linkedin-fmt style "Shipped v1.0.2" --family script
𝒮𝒽𝒾𝓅𝓅ℯ𝒹 𝓋1.0.2

$ linkedin-fmt style "Shipped" --bold
𝐒𝐡𝐢𝐩𝐩𝐞𝐝

$ linkedin-fmt mark "Wrong" --mark strikethrough
W̶r̶o̶n̶g̶

Install

npm install -g linkedin-fmt

Or run it without installing:

npx -y linkedin-fmt --help

Node 18 or newer. No dependencies, one file, nothing to configure.

Commands

| Command | What it does | |---|---| | style <text> | Apply a font family, optionally bold and/or italic | | mark <text> | Layer a combining mark (strikethrough, underline) | | strip <text> | Convert styled text back to plain ASCII | | count <text> | Count against LinkedIn's 3,000 character limit | | families | List the 11 families and the axes each supports | | styles | List every letterform style and combining mark |

Options

| Flag | For | Meaning | |---|---|---| | -f, --family <id> | style | Font family (default serif) | | -b, --bold | style | Bold axis, where the family has one | | -i, --italic | style | Italic axis, where the family has one | | -m, --mark <id> | mark | strikethrough or underline | | -u, --unit <unit> | count | utf16 (default) or codepoints |

Input

Text comes from the argument, or from stdin when the argument is omitted, so files work by redirection and commands compose:

$ linkedin-fmt strip < post.linkedin
$ cat post.linkedin | linkedin-fmt count
$ linkedin-fmt style "Hello" --family fraktur | linkedin-fmt strip
Hello

Counting

LinkedIn counts UTF-16 code units, not characters. A styled character costs two, so a naive character count will tell you a post fits when it does not:

$ linkedin-fmt count "Hello"
5 / 3000  [utf16, normal]

$ linkedin-fmt style "Hello" --bold | linkedin-fmt count
10 / 3000  [utf16, normal]

count exits non-zero when over the limit, so a publishing script can gate on it:

linkedin-fmt count < post.linkedin || { echo "too long"; exit 1; }

Behaviour worth knowing

Fail closed. A character with no equivalent in the target style is left unchanged, never swapped for a visual lookalike. Families with no digit row leave digits as plain ASCII. That is correct rather than a fallback.

Unicode is sparse. Monospace has no bold; script has no italic. Asking for an axis a family does not have drops it, reports that on stderr, and leaves stdout clean so pipelines keep working:

$ linkedin-fmt style "Hi" --family monospace --bold
𝙷𝚒
note: monospace has no bold in Unicode; dropped.

Lossless round trip. strip reverses style exactly, for every family, asserted over the full A-Za-z0-9 range.

No network. Nothing is sent anywhere. There is no telemetry and no update check.

Related

• VS Code extension - the same core with a live LinkedIn card preview: code --install-extension kaushiknaru.linkedin-formatter
• MCP server - the same core as tools for AI assistants: claude mcp add -s user linkedin-formatter -- npx -y linkedin-formatter-mcp, or npx -y linkedin-formatter-mcp from any other client's config

All three share one conversion core with no duplicated logic.

Licence

MIT