Clean up unicode, whitespace, and encoding artifacts in text before LLM input via @mukundakatta/textsanity-mcp (npx)

Recipe: @mukundakatta/textsanity-mcp v0.1.1

Install: npm install @mukundakatta/textsanity-mcp Transport: stdio Tools: 2 — sanitize, normalize_newlines

Tool: sanitize

Clean up unicode/whitespace in a string. Params: text (required), plus optional flags: preset ("default"|"strict"), nfkc, strip_zero_width, strip_control, collapse_whitespace, trim, ascii_punctuation, strip_emoji, ascii_only.

Tool: normalize_newlines

Convert CRLF and CR to LF. Params: text (required).

Execution trace (8 calls, 100% success)

Key gotchas

1. Default mode is VERY conservative — zero-width chars, control chars, and excessive whitespace are NOT stripped unless you pass explicit flags or preset: "strict". The description says defaults are on but behavior suggests otherwise.
2. `preset: "strict"` is aggressive — strips ALL non-ASCII (é→nothing, £→nothing, ¥→nothing), not just problematic chars. Use individual flags for finer control.
3. `ascii_punctuation: true` is the most useful flag — converts curly quotes, em dashes, and ellipses to ASCII equivalents without destroying content.
4. `normalize_newlines` is independent from sanitize — call separately.
5. First call ~40ms (JIT), subsequent 4-30ms.
6. Output is always JSON {"clean": "..."}.