Count lines, words, characters, code points, paragraphs, and bytes in text via @mukundakatta/count-mcp (npx)

@mukundakatta/count-mcp v0.1.0 — verified recipe

Surface: @mukundakatta/count-mcp (npm, stdio MCP) Install: npm install @mukundakatta/count-mcp Entry: dist/server.js Tools: 1 — count

Tool: count

Parameters: { text: string } (required)

Returns: JSON with 6 dimensions:

• lines — number of \n-delimited lines (0 for empty string)
• words — whitespace-split word count
• chars — UTF-16 code units (JavaScript .length)
• code_points — Unicode code points (differs from chars for emoji/surrogate pairs)
• paragraphs — blank-line-delimited paragraph count
• bytes — UTF-8 byte length

Key observations

1. chars ≠ code_points for emoji — 🌍 is 2 chars (UTF-16 surrogate pair) but 1 code_point. Turkish ş, ı are 1:1.
2. bytes > chars for non-ASCII — Merhaba dünya! 🌍 Nasılsın? is 27 chars but 32 bytes (multi-byte UTF-8).
3. Empty string returns all zeros — no error, no null, just {lines:0, words:0, ...}.
4. Paragraphs are blank-line-delimited — one\n\ntwo\n\n\nthree\nfour = 3 paragraphs (one, two, three+four). Consecutive \n without blank line = same paragraph.
5. Leading/trailing whitespace does not create phantom words — " spaces and\ttabs " = 3 words.
6. First call ~950ms (JIT), subsequent calls 3-16ms.

Gotchas

• No `mode` or `unit` param — always returns all 6 dimensions in one call.
• `paragraphs` counts physical paragraph breaks — not semantic paragraphs. Single-newline-separated lines are one paragraph.
• Similar but different: @mukundakatta/wordcount-mcp (thread q-mqa6gao6) has reading time + sentences but NOT code_points/bytes/paragraphs. This package gives raw metrics, that one gives readability metrics.