S3eidat Tools

Render a chart to an SVG or PNG file, offline. Either give a simple description ({ type: bar|line|area|pie|scatter, data: rows, x, y, color?, title? }; y may be a list of fields for multi-series line/bar charts), or a full Vega-Lite (or Vega) spec in "spec" with the rows inline in data.values. Field types are inferred (numbers = quantitative, ISO dates = temporal, other = categories). Loading data or images from URLs or files is refused. Up to 50,000 rows. PNG is rasterised from the SVG at "scale" (default 2×).

Split text, markdown or a text file into token-sized chunks for retrieval (RAG) and embeddings. Splits at markdown headings (ATX "#" headings, never inside code fences) and paragraphs first, then sentences (lines inside code blocks), and only cuts mid-sentence when a single sentence is longer than maxTokens. Consecutive chunks of the same section share about overlapTokens of text. Each chunk returns index, text, tokens, startChar/endChar (JavaScript string offsets, end exclusive, so text === source.slice(startChar, endChar) once CRLF line endings are normalised to \n) and the heading path it sits under. Tokens are counted with tiktoken (o200k_base by default; for Claude treat counts as estimates). Give "text" or "path" (file path or base64); give "output" to also write the chunks as JSONL (one JSON object per line).

Build a 16:9 PowerPoint (.pptx) deck from a simple outline: an optional title slide, then slides with a title plus any of bullets (nested), a paragraph, an image (file path or base64), a table (first row = header; long tables continue on extra slides), a native editable chart (bar, line, pie) and speaker notes. Light or dark theme with clean, readable defaults; text plus a visual are laid out side by side. Arabic/Hebrew text is set right-to-left. Up to 300 slides. Opens in PowerPoint, Keynote, Google Slides and LibreOffice.

Build a real Excel .xlsx file from data: one or more sheets, each with optional column definitions (header, key, width, numFmt) and rows given as objects or arrays. Bold shaded header, frozen header row and filter buttons (both on by default), column widths fitted to the content, Excel number formats (raw codes like "#,##0.00" or aliases: integer, number, percent, currency, usd, eur, gbp, date, datetime, time, text), strings starting with "=" written as live formulas (=SUM(B2:B9), =B2*C2, cross-sheet refs), ISO date strings written as real dates, and an optional totals row (SUM/AVERAGE/COUNT/MIN/MAX formulas). Up to 50 sheets and 1,000,000 rows per sheet. Formulas are calculated by Excel when the file is opened. Writes to "output" (default "workbook.xlsx" in the output folder) and returns the path and a summary of each sheet.

Profile a table to understand and clean it before analysis: row and column counts, and per column the inferred type (integer, number, boolean, date, datetime, string), non-null and null counts, unique count, top 5 values, sample values, min/max/mean/median/stddev for numbers, min/max for ISO dates, min/max length for text, plus data-quality warnings (mixed types, leading/trailing spaces, mostly missing, constant columns, duplicate rows, ragged headers). Input: "file" (local path or base64 of CSV, TSV, JSON array, JSONL or .xlsx, first sheet or "sheet") or inline "csv" / "json". Empty cells, "null" and "NULL" count as missing. Runs locally; nothing is uploaded.

Detect the language of a text with franc-min (82 languages with 8M+ speakers, ISO 639-3 codes) and the writing systems it uses (Latin, Arabic, Cyrillic, Han/CJK, Hangul, Devanagari, Greek, Hebrew, Thai…). Returns the top 3 candidates with scores (1 = best match, relative to the others), English names and ISO 639-1 codes where they exist, plus a share per script. Texts under 20 characters are "too short": only the script is reported. Works best on a sentence or more of one language; for mixed text the dominant script wins; long texts are judged on their first 100,000 characters. Hebrew is not among the languages (script only). Give "text" or "path" (file path or base64).

Draw a flowchart / graph diagram from a tiny text language and save it as SVG or PNG, laid out automatically (dagre). One statement per line: `direction: LR` (TB, LR, BT, RL), `title: My flow`, `A [shape=diamond, label="Ready?", color=#fde68a]` (shapes: box, round, diamond, circle, ellipse), edges `A -> B: label`, chains `A -> B -> C`, `-->` dashed, `==>` thick, `<->` both ends, `--` no arrow; quote names with spaces ("Start here" -> A); `\n` in labels breaks lines; # starts a comment. Or pass JSON { nodes: [{id, label, shape, color}], edges: [{from, to, label, style, arrow}] } as "graph". Up to 500 nodes and 2000 edges. Labels use Helvetica/Arial metrics (Latin text is measured best).

Convert a Word .docx document to Markdown: headings (from Heading styles), bold/italic, lists, links, tables, footnotes. Images become placeholders like  unless embedImages is true (then data: URIs, which can be very large). Returns the number of conversion warnings (unrecognised styles and the like) and the first few messages. Old binary .doc files are not supported. Accepts a local path or base64. Inline Markdown is cut at maxChars (default 200k); "output" saves it to a file. Runs locally; nothing is uploaded.

Pull structured items out of text or a text file, deduplicated with counts: emails, URLs, phone numbers (libphonenumber, validated, E.164 plus type/country; local numbers read with defaultRegion, default JO), dates (ISO, D/M/Y or M/D/Y, Y/M/D, English month names; normalised to YYYY-MM-DD; a date like 03/04/2024 is flagged ambiguous with the other reading), money ($, €, £, JOD/JD/د.أ, SAR/ر.س, AED/د.إ, KWD, EGP… and Arabic currency words, with amount and ISO currency), percentages, #hashtags, @mentions, IPv4/IPv6, IBANs (country length and mod-97 checked) and card-like numbers (only Luhn-valid ones with a known brand prefix, always masked: only the last 4 digits are shown, the full number is never returned). Arabic-Indic digits are understood. Regex-based, not NLP: no names, places or organisations. Give "text" or "path" (file path or base64).

Find out what a file really is from its content (magic bytes, via file-type: images, video, audio, archives, Office, PDF, fonts, executables… ~150 formats), not its name. Reports the detected type and MIME, whether the extension disagrees with the content, size, SHA-256, and for text files an encoding guess (UTF-8, UTF-8/16/32 with BOM, UTF-16 without BOM, ASCII, 8-bit legacy) plus a text kind (JSON, XML, SVG, HTML…). Accepts a file path or base64. Nothing is executed.

Convert HTML to clean Markdown for reading or feeding to a model: headings, lists, links, images, code blocks, GitHub-style tables and strikethrough. Give the HTML inline ("html") or as a file ("path": local path or base64). Optional "readable" keeps only the main article (Mozilla Readability, falls back to the whole body), "selector" converts only matching elements, "baseUrl" makes relative links absolute, "includeLinks": false drops link URLs and images. Scripts, styles and forms are removed. Inline Markdown is cut at maxChars (default 200k) with a note; give "output" to save the whole thing to a .md file. Runs locally; nothing is uploaded.

Compare two images pixel by pixel (pixelmatch, anti-aliasing aware) and write a diff PNG with changed pixels in red over a faded copy of the first image. Returns the number and percentage of different pixels. Images must be the same size unless resize is true (the second is stretched to the first's size). Reads PNG, JPEG, WebP, AVIF, GIF (first frame), TIFF and SVG; file path or base64; up to 50 MP. Good for visual regression checks.

Generate a JSON Schema (draft 2020-12) from one or more sample JSON values: types (integer vs number, mixed types as a type list), "null" added where a value can be null, required = keys present in every sample object, array items merged across all elements, string formats guessed when every sample matches (date-time, date, email, uri, uuid), and optional enums for small repeated string sets. Give "samples" (an array of sample values), or "text" (JSON, or JSONL = one sample per line), or "file" (path or base64 of a .json/.jsonl file). An array sample describes an array; to treat its elements as separate samples set eachItem. Pure local code; optionally writes the schema to "output".

Validate JSON data against a JSON Schema with Ajv, reporting every error (allErrors) with its instancePath (JSON Pointer), schemaPath, keyword, message and params. Draft 2020-12 is used when the schema's $schema says 2020-12 (2019-09 likewise), otherwise draft-07 (draft-04/06 schemas are checked as draft-07). Formats checked: date, time, date-time, email, hostname, ipv4, ipv6, uri, uri-reference, uuid, regex, json-pointer (other formats are ignored and listed). Data: inline "data" (any JSON value), "dataText" (JSON text) or "dataFile" (path or base64). Schema: inline "schema" (object or JSON text) or "schemaFile". Remote $ref URLs are never fetched. Runs locally.

Convert Markdown (GitHub flavoured) to a real Word .docx: headings H1-H6 (Word heading styles, so the navigation pane and table of contents work), paragraphs with bold, italic, strikethrough, inline code and clickable links, bullet and numbered lists (nested, numbering restarts per list, task-list checkboxes), tables (header row shaded, column alignment kept), blockquotes, fenced code blocks in a monospace font, horizontal rules, and a page break wherever a line says \newpage. Arabic/Hebrew paragraphs are laid out right-to-left automatically (rtl: "auto"). Images are not embedded (their alt text is shown); raw HTML tags are dropped. Options: title and author (document properties), font (default Calibri), fontSize. Give "markdown" inline or "path" to a .md file. Writes the .docx to "output" (default name from the title or first heading).

Turn Markdown (GitHub-flavoured) into a clean, paginated A4/Letter/A5 PDF: headings, paragraphs with word wrap, bold/italic/strikethrough, inline code, links (clickable), bullet / numbered / task lists (nested), tables (wrapped cells, header repeated on new pages), blockquotes, code blocks, horizontal rules, local images (relative to the Markdown file; remote images are never fetched and show as alt text), page breaks (\pagebreak, \newpage or <!-- pagebreak -->), page numbers, margins and title metadata. Text uses the built-in PDF fonts: Latin text only (English and Western European). Arabic, CJK and other scripts are not supported yet: such characters are replaced with "?" (or rejected with onUnsupportedChar: "error"); emoji are dropped; arrows and common symbols are transliterated. Give "markdown" inline or "path" to a .md file.

Read audio and video file metadata offline (music-metadata): container, codec, duration, bitrate, sample rate, bit depth, channels, lossless or not, tags (title, artist, album, year, genre, track…), embedded cover art, and per-track info (including video width/height for MP4/MKV/WebM). Formats: MP3, AAC/M4A/MP4/MOV, FLAC, WAV, AIFF, OGG/Opus/Vorbis, WebM/Matroska, WMA/ASF, APE, WavPack, DSD and more. File path (streamed, any size) or base64. Duration of VBR MP3s without a header is computed by scanning the file.

Read the text in an image (PNG, JPEG, WebP, TIFF incl. multi-page, GIF, AVIF) or a scanned PDF with Tesseract (tesseract.js), fully offline: the English and Arabic models ship with the server and nothing is downloaded or uploaded. Languages: "eng", "ara" or "eng+ara" (default). PDF pages are rendered at 200 dpi by default. Returns the text per page and the mean word confidence (0-100); give "output" to also write a .txt file. Limits: up to 200 pages per call, roughly 1-5 s per page; printed text works well, handwriting and very small or skewed text poorly. For PDFs that already have a text layer, pdf_to_text is faster and exact. Give "file" as a local path or base64.

Convert a text PDF to structured Markdown: headings guessed from font size (text clearly larger than the body becomes #, ##, ###), paragraphs rebuilt from line spacing with hyphenated words mended, bullet and numbered lists kept, right-to-left lines in reading order. Optional page range ("1-5, 8") and page markers. Scanned PDFs without a text layer give empty pages (listed in pagesWithoutText; run OCR first). Tables and columns are read line by line. Accepts a local path or base64. Inline Markdown is cut at maxChars (default 200k); "output" saves it to a file.

Extract the text of a PowerPoint .pptx deck as Markdown: one "## Slide N: title" section per slide in presentation order, text boxes as paragraphs and bullet lists (with indent levels), tables as Markdown tables, and speaker notes as a quote under each slide. Images, charts and SmartArt text are not extracted. Old binary .ppt is not supported. Optional "slides" range (e.g. "1-5, 8"). Accepts a local path or base64. Inline Markdown is cut at maxChars (default 200k); "output" saves it to a file. Runs locally; nothing is uploaded.

Run one read-only SQL SELECT (alasql dialect: JOIN, GROUP BY, HAVING, ORDER BY, LIMIT, UNION, WITH, subqueries, CASE, SUM/COUNT/AVG/MIN/MAX/MEDIAN/STDEV, string and date functions) over one or more tables. Each table is { name, file | csv | json }: a local path or base64 of a CSV, TSV, JSON array, JSONL or .xlsx (first sheet, or "sheet"), or inline CSV text / JSON rows. CSV cells are typed (numbers, true/false, empty → null; ISO dates stay ISO text, so they sort and compare correctly; YEAR()/MONTH() work on them). Only SELECT is allowed: INSERT/UPDATE/DELETE/DROP/CREATE/ALTER/ATTACH, SELECT INTO, file readers like FROM CSV(...), inline JavaScript and the -> operator are refused. Table names are case-sensitive; quote odd column names and reserved words (total, value, count, date…) as [My Column]. Returns columns and up to maxRows rows (default 500) with the total count; set "output" (.csv, .tsv, .json, .jsonl or .xlsx) to save every row to a file. Runs locally; nothing is uploaded.

Compare two or more texts (or text files) for overlap. For every pair returns: Jaccard similarity of word shingles (k=3 words by default), a MinHash estimate of that Jaccard (128 hashes), and cosine similarity of word frequencies. Words are lower-cased, punctuation and Arabic diacritics ignored. Measures wording overlap, not meaning (no embeddings): paraphrases score low. Mode "pairs" (default) lists pairs sorted by the chosen metric; mode "dedupe" groups near-duplicates whose metric is at least "threshold" (default 0.8; groups are connected, so A~B and B~C put A, B, C together). Give "texts" (inline strings) and/or "paths" (file paths or base64). Up to 1000 texts.

Count LLM tokens in a text or text file with the OpenAI tiktoken tokenizers ("o200k_base", default, used by GPT-4o and the o-series; or "cl100k_base", used by GPT-4/3.5 and text-embedding-3). Also returns characters and words, and whether the text fits in common context windows (8k, 32k, 128k, 200k, 1M: e.g. "fits in 200k: yes"). Claude's tokenizer is not public, so for Claude the count is an estimate (typically within ~10-20%; Claude usually needs a little more), returned as a rough range. Runs locally, no API calls. Give "text" or "path" (file path or base64). Large texts (many MB) take a few seconds.

Fetch a web page (http/https) and return its readable content as clean Markdown, with title, byline, final URL after redirects and a word count. Uses Mozilla Readability to keep the main article (falls back to the whole page), or a CSS "selector" you give. Respects robots.txt (refuses pages disallowed for the s3eidat-tools-mcp user agent), only reads HTML and text responses, stops at 5 MB and 20 seconds, follows up to 5 redirects, and refuses private, local and loopback addresses unless allowPrivate is true. Does not run JavaScript, so pages that build their content in the browser may come back nearly empty. Inline Markdown is cut at maxChars (default 200k); "output" saves it to a file.

Turn an Excel .xlsx workbook (or a .csv/.tsv file) into Markdown tables, one "## sheet" section per sheet, or only the sheet you name. The first used row is the table header; empty rows and columns around the data are dropped. Formulas show their cached result (as last saved by Excel), dates are ISO (2026-10-30, or 2026-10-30T14:05:00), merged cells show their value once, hyperlinks become Markdown links. Each sheet shows at most maxRows data rows (default 200) and says how many were left out. Old binary .xls is not supported. Accepts a local path or base64. Runs locally.

Work with ZIP archives locally. action "create": zip files and folders (given as paths; folders keep their own name as the top folder, symlinks are skipped) into one .zip. action "extract": unpack a .zip (path or base64) into the "output" folder; entries with "..", absolute paths or drive letters are refused (zip-slip protection), existing files are kept unless overwrite is true. action "list": entry names with sizes, no extraction. Limits: total uncompressed size 500 MB and 10000 files by default (maxTotalMB / maxFiles). Encrypted ZIPs are not supported.

Make 1D barcodes as SVG or PNG: Code 128 (any printable ASCII; digit runs packed with subset C automatically), EAN-13 (12 digits, check digit added; or 13, check digit verified), UPC-A (11 or 12 digits, same) and Code 39 (A-Z, 0-9, space, - . $ / + %, optional mod-43 check). Same encoders, quiet zones, guard bars and human-readable digits as the website. Give "value" for one barcode, or "values" (up to 1000) to write one file per value into a folder; invalid values are reported, not fatal. PNG text uses the system monospace font. Web version: https://s3eidat.github.io/barcode-gen/

Convert a whole number of any size (BigInt) between bases 2–36: binary, octal, decimal, hex, base 36 or any other. Input base can be auto-detected from 0x / 0b / 0o prefixes; spaces, _ and , separators are ignored; negatives OK. Also shows bit length and the two's complement view (signed/unsigned value, hex, bits) at 8/16/32/64 bits, and whether it fits. Fractions are not supported (same as the website). Web version: https://s3eidat.github.io/base-convert/

Decode Base64 (standard or URL-safe, padding optional, whitespace and data: URI prefix allowed). Valid UTF-8 text is returned inline; binary data (or anything when "output" is given) is written to a file whose extension comes from the detected type (PNG, JPEG, GIF, WebP, BMP, PDF, ZIP, gzip, MP3, Ogg, MP4, WOFF2, SVG, JSON). Web version: https://s3eidat.github.io/base64-kit/

Encode text (as UTF-8) or a file (path or base64) to Base64. Options: URL-safe alphabet (- and _, no padding), wrap at 76 columns (MIME), or a data: URI for files (MIME type detected from magic bytes). Results over 1 MB, or when "output" is given, are written to a text file instead of returned inline. Web version: https://s3eidat.github.io/base64-kit/

Turn text into its UTF-8 bytes (hex, binary or decimal), or bytes back into text. Byte input is lenient: "48 65", "4865", "0x48,0x69", "01001000 01101001", "72 105". Invalid UTF-8 is reported and shown with � replacements. Web version: https://s3eidat.github.io/base-convert/

Exact age in years, months and days from a date of birth (on today or any "on" date), with totals in months, weeks, days and hours, the weekday of birth, Hijri (Umm al-Qura) birth date and Hijri age, next birthday (Feb 29 birthdays fall on Feb 28 in common years), star sign, Chinese zodiac, the next 1,000-day milestone, and optionally the age gap to a second person. Dates are ISO YYYY-MM-DD. Web version: https://s3eidat.github.io/age-calc/

Body mass index with the WHO adult category, the healthy weight range for your height (BMI 18.5–24.9), BMI Prime, optional waist-to-height ratio, and for adults (18–120, with age) BMR by Mifflin-St Jeor plus daily calories to maintain or lose ~0.5 kg (1 lb) a week at five activity levels. Metric (cm, kg) or imperial (ft + in, lb). Height 50–272 cm, weight 10–650 kg. Not medical advice; children need percentile charts. Web version: https://s3eidat.github.io/bmi-calc/

Count a cash drawer by denomination with exact integer maths (no float errors): JOD (3 decimals, fils), USD, EUR, SAR, AED, GBP, KWD, EGP, or a custom currency (your own note and coin values, 0/2/3 decimals). Give how many of each note/coin (by value like 50 or 0.25, by label like "50 JD" or "¼ dinar"); returns per-denomination amounts, the total, notes and coins counted, and over/short against an expected amount. list=true shows the denominations of a currency. Web version: https://s3eidat.github.io/cash-counter/

Project savings with compound interest: an initial deposit plus monthly contributions (at the start or end of each month), a nominal annual rate compounded yearly, half-yearly, quarterly, monthly, daily or continuously, optionally adjusted for inflation ("today's money"). Returns final balance, total deposits, interest earned and a year-by-year table. Rate -50–100 %, 1 month to 100 years. Web version: https://s3eidat.github.io/compound-interest/

Two date jobs. difference: days between two dates (optionally counting the end day), as years/months/days, weeks, whole months, working days and weekend days for a chosen weekend (Fri–Sat default, Sat–Sun, Fri only, Sun only; public holidays are not counted). add: add or subtract days, business days, weeks, months or years to a date (month ends clamp: Jan 31 + 1 month = Feb 28/29). Results include the weekday and Umm al-Qura Hijri date. Dates are ISO YYYY-MM-DD; for Hijri ⇄ Gregorian conversion use calc_hijri. Web version: https://s3eidat.github.io/date-diff/

Three shop maths jobs. sale: price after one or two stacked discounts (each a percent or a fixed amount) plus optional tax/VAT, with what you save and the effective total discount (stacked percents multiply). reverse: the discount % between an old and a new price. compare: which of two deals is cheaper per unit (price and amount of each). Money is rounded to the cent at each step like a till. Web version: https://s3eidat.github.io/discount-calc/

Estimated pregnancy due date from the first day of the last period (Naegele's rule, adjusted for cycle length 20–45 days), the conception date, or an IVF embryo transfer date (day-3 or day-5 embryo). Also gives the estimated conception date, full-term date, how far along you are on a given day (weeks + days, trimester, days to go) and a dated timeline of common milestones (weeks 4–42), optionally with Hijri dates. Estimate only, not medical advice. Web version: https://s3eidat.github.io/due-date/

Solve equations exactly with worked steps, like the website: linear a·x + b = c, quadratic a·x² + b·x + c = 0 (exact rational or surd roots like 1/2 + √5/2, complex roots), cubic a·x³ + b·x² + c·x + d = 0 (rational root test, else trigonometric/Cardano to 10 significant digits), and systems of 2 or 3 linear equations (Gauss–Jordan; no solution and infinitely many solutions are detected and parametrised). Coefficients may be integers, decimals, fractions ("3/4"), mixed numbers ("-1 1/2") or e-notation. Web version: https://s3eidat.github.io/equation-solver/

Evaluate a maths expression safely with the website's own parser (never eval): + − × ÷ ^ (right-associative), implicit multiplication (2(3+4), 2x), postfix ! and %, √, brackets; functions sin cos tan sec csc cot (+ asin acos atan atan2) in degrees (default), radians or grads, sinh cosh tanh and inverses, sqrt cbrt root(x,n) abs exp ln log(x[,base]) log10 log2 floor ceil round sign min max mod nCr nPr fact gamma; constants pi π e tau phi; "ans" and your own variables. Results to 12 significant digits, plain or scientific notation. Web version: https://s3eidat.github.io/sci-calc/

Exact fraction maths with big integers (never rounded) and worked steps. calculate: an expression with fractions, mixed numbers ("1 1/2"), decimals and repeating decimals ("0.(3)"), + − × ÷ ^ and brackets (a/b binds tighter than ÷). decimal_to_fraction: "0.75", "0.1(6)" (brackets mark the repeating part), with close small-denominator approximations. fraction_to_decimal: exact decimal with the repeating block, and percent. compare: put two or more fractions in order. Web version: https://s3eidat.github.io/fraction-calc/

Fuel needed and cost of a trip from distance (km or miles, optionally round trip), fuel price (per litre, US gallon or UK gallon) and consumption in L/100 km, km/L, mpg (US) or mpg (UK); split between people; cost per km/mile and per 100. Optionally compare a second car (its own consumption and, if different, fuel price). Also converts the consumption to the other units. Web version: https://s3eidat.github.io/fuel-cost/

Credit-weighted semester GPA, cumulative GPA (with previous credits and GPA) and the average you need over the remaining credits to reach a target. Scales: 4.0 (A…F, no A+), 4.3 (A+ = 4.3), 5.0 (A+ 5, A 4.75, B+ 4.5 … F 1) with letter grades, or "percent" with numeric marks between a minimum and maximum (default 35–100). Courses with no credits or an invalid grade are skipped (and listed). Up to 40 courses. Web version: https://s3eidat.github.io/gpa-calc/

Convert between Gregorian and Hijri dates with the Umm al-Qura calendar (the same Intl calendar the website uses), both ways. Modes: to_hijri (a Gregorian date), to_gregorian (Hijri year, month 1–12, day), month (length of a Hijri month and its Gregorian start/end), age (age in Hijri and Gregorian years, next Hijri birthday). Results come in English and Arabic (Arabic-Indic digits by default; Gregorian month names in the Egyptian/Gulf or Levant style). Supported range 1300–1600 AH (1882–2174). Web version: https://s3eidat.github.io/hijri-converter/

Weekly time card: add shifts (start and end "HH:MM", unpaid break minutes; an end at or before the start is an overnight shift), up to 6 shifts a day. Totals per shift, per day and per week in h:mm and decimal hours; optional rounding of each shift to 5, 6 or 15 minutes; overtime over N hours a day and/or N regular hours a week (daily first, then weekly), paid at a multiplier (default 1.5); pay at an hourly rate. Web version: https://s3eidat.github.io/hours-calc/