The same utilities, inside your assistant.

Add it to a client

Clone the repository, then point your client at mcp/server.mjs. Node.js 24 or newer is required.

{
  "mcpServers": {
    "filegizmo": {
      "command": "node",
      "args": ["/absolute/path/to/filegizmo/mcp/server.mjs"]
    }
  }
}

Restart the client afterwards. It should list decode_jwt, merge_record_sets, and parse_page_ranges among its tools. From a checkout you can also run npm run mcp to start the server directly and confirm it responds.

42 tools

Every call is deterministic and read-only. Inputs and outputs stay between your client and this local subprocess.

• text_statsCount words, lines, graphemes, UTF-16 units, and reading time.
• convert_caseConvert text to upper, lower, sentence, title, camel, or snake case.
• remove_duplicate_linesRemove repeated lines while optionally trimming and ignoring case.
• slugifyCreate a lowercase URL or filename slug.
• generate_loremGenerate bounded Lorem ipsum paragraphs.
• diff_linesCompare two strings line by line.
• smart_csv_valueCoerce one CSV value without corrupting phone numbers, leading zeros, or long identifiers.
• smart_csv_recordsApply conservative CSV typing to an array of records.
• flatten_recordFlatten a nested record into dot-separated keys.
• normalize_json_recordsNormalize JSON values into compact or flattened tabular records.
• merge_record_setsMerge two or more parsed CSV record sets by strict or union headers.
• format_jsonFormat JSON with optional key sorting.
• decode_jwtDecode a JWT header and payload without verifying its signature.
• developer_error_messageMap a developer-tool exception message to safe user-facing copy.
• json_fidelity_noticesDetect unsafe large integers and duplicate keys before formatting JSON.
• parse_page_rangesParse ranges such as 1-3,7 into zero-based page indexes.
• format_bytesFormat a byte count in a human-readable binary unit.
• encode_base64Encode text as base64, optionally in the URL-safe alphabet.
• decode_base64Decode standard or URL-safe base64 back to text.
• encode_url_textPercent-encode text as a URL component or as a full URL.
• decode_url_textDecode percent-encoded text from a URL component or a full URL.
• generate_uuidsGenerate cryptographically random version 4 UUIDs.
• generate_passwordGenerate one random password and report its entropy in bits.
• password_entropyScore a password policy in bits of entropy without generating one.
• hash_textHash text with SHA-256, SHA-384, SHA-512, or SHA-1.
• timestamp_reportRead a Unix timestamp or date string and lay out every scale and zone.
• looks_like_formulaDetect a CSV value a spreadsheet would execute as a formula.
• dedupe_rowsRemove duplicate CSV records, either whole-row or by key columns.
• split_rowsSplit records into chunks of a fixed size or into a fixed number of parts.
• extract_columnsKeep only the chosen columns from a set of CSV records.
• csv_rows_to_htmlRender CSV records as an accessible HTML table.
• geocode_estimateEstimate how long geocoding a number of rows would take.
• qr_svgBuild a QR code and return it as a standalone SVG string.
• pdf_infoRead a PDF page count, page sizes, rotations, and document metadata without changing it.
• pdf_to_textRead the text out of a PDF, page by page, without changing the file.
• pdf_mergeJoin several PDFs into one, in the order given, and write it to disk.
• pdf_splitSplit one PDF into separate files by page ranges such as 1-3,7 and write them into a directory.
• pdf_extract_pagesCopy the named pages of a PDF into a new PDF.
• pdf_delete_pagesRemove the named pages from a PDF and write the rest to a new PDF.
• pdf_rotate_pagesTurn the named pages of a PDF by 90, 180, or 270 degrees, added to the rotation they already carry.
• pdf_reorder_pagesRewrite a PDF with its pages in a new order, given as every zero-based position exactly once.
• wifi_payloadBuild the WIFI payload string a phone camera joins a network from.

What it deliberately does not do

• No browser-only work. Compressing a PDF, removing an image background, and running OCR need DOM, Worker, WASM, or pdf.js APIs. They stay in the browser rather than being reimplemented here, because a second implementation is a second set of results to keep honest.
• No signature verification. decode_jwt reads a header and payload and does not check that the token is genuine. Treat the claims as untrusted until your own code verifies them.
• No writes. Nothing on this server creates, edits, or deletes a file on your disk.
• Indexes, not page numbers. parse_page_ranges returns zero-based indexes for code. The page a reader calls 1 is index 0.