Claude skill

The S3eidat Tools skill

skills/s3eidat-tools/SKILL.md teaches an AI assistant which S3eidat tool fits which job, how to chain tools, the privacy promise, and when to send a person to the web version instead. It works together with the MCP server (npx -y s3eidat-tools-mcp): install the server first (see the main README), then the skill.

Claude Code

Personal (all projects):

mkdir -p ~/.claude/skills
cp -r skills/s3eidat-tools ~/.claude/skills/

Or for one project: copy the folder to .claude/skills/s3eidat-tools/ in that project. Without cloning the repo:

mkdir -p ~/.claude/skills/s3eidat-tools && curl -fsSL https://raw.githubusercontent.com/s3eidat/s3eidat-tools-mcp/main/skills/s3eidat-tools/SKILL.md -o ~/.claude/skills/s3eidat-tools/SKILL.md

The npm package also ships the skill: cp -r "$(npm root -g)/s3eidat-tools-mcp/skills/s3eidat-tools" ~/.claude/skills/ after npm i -g s3eidat-tools-mcp.

Claude.ai and Claude Desktop

Download the folder as a zip: skills/s3eidat-tools/ must be the top folder inside the zip (cd skills && zip -r s3eidat-tools.zip s3eidat-tools).
In Claude: Settings → Capabilities → Skills → Upload skill, and pick the zip.
Make sure the S3eidat Tools MCP server is connected in Claude Desktop (claude.ai on the web can't reach a local server, so there the skill mainly helps Claude point people to the right web tool).

Other assistants

Most agents that read Markdown instructions can use SKILL.md as-is: add it to the agent's rules/instructions file (for example AGENTS.md, Cursor rules, or a Copilot instructions file).

The skill itself

Source: skills/s3eidat-tools/SKILL.md

--- name: s3eidat-tools description: Use the S3eidat Tools MCP server (tools named pdf_, image_, text_, json_, csv_, calc_, qr_, doc_ …) for everyday file and data chores done locally and privately — merging, splitting, compressing, rotating, watermarking, numbering, signing or filling PDFs; PDF to text or images; images to PDF; converting, resizing, cropping, compressing images, reading or stripping EXIF/GPS; word counts, case, slugs, diffs, Arabic text fixes, numbers to words; formatting or converting JSON/CSV/YAML/XML/SQL; hashes, Base64, JWT, UUIDs, regex, cron, timestamps, subnets; loan, interest, percentage, zakat, GPA, BMI, date and unit calculations, Hijri dates; prayer times and qibla; passwords and AES encryption; QR codes, barcodes; invoices, receipts, labels, calendars and CVs as PDF. Trigger whenever the user asks for one of these jobs, even without naming the tools. ---

S3eidat Tools

A local MCP server (npx -y s3eidat-tools-mcp) with about 140 small, reliable tools (plus extra tools made for AI agents) ported from https://s3eidat.github.io. Everything runs on the user's machine: files are never uploaded. Prefer these tools over writing ad-hoc scripts for the same job: they are tested against the website's own known answers, so results match.

How to call them

File inputs take a local path (absolute, ~/…, or relative to the server's --root) or base64. Prefer paths: they are faster and keep big files out of the conversation.
File outputs go to the output you pass (a file path, or a folder ending in /). Without one they go to a temp folder and the result tells you the path. Always tell the user where the file is.
Results are JSON with small metadata (path, bytes, page counts…). Ask for returnBase64 only when you really need the bytes in the conversation.
Errors come back as clear messages (isError). Fix the input and retry; don't loop on the same call.
If the server was started with --root, every path must be inside that folder.

Compact mode

If the server was started with --compact, you will only see s3eidat_find_tools, s3eidat_tool_info and s3eidat_run_tool. Find the tool by keywords (e.g. "merge pdf"), read its inputs with s3eidat_tool_info, then call s3eidat_run_tool with name and arguments. The tool names below are the same in both modes.

Which tool for which job

Job	Tool(s)
Page count, sizes, encryption, text layer?	`pdf_info` (a good first call before other PDF work)
Combine PDFs	`pdf_merge`
Split / pull out pages	`pdf_split`, `pdf_extract_pages`
Reorder, delete, rotate pages	`pdf_organize`, `pdf_rotate`
Make a PDF smaller	`pdf_compress`
Stamp text/logo, add page numbers	`pdf_watermark`, `pdf_page_numbers`
Read or clean PDF metadata	`pdf_metadata_read`, `pdf_metadata_write`
Get the text out of a PDF	`pdf_to_text`, or `ai_pdf_to_markdown` to keep headings and lists (scanned PDFs have no text: use `ai_ocr`)
Photos/scans → one PDF	`pdf_from_images`
PDF pages → PNG/JPEG	`pdf_to_images`
Put a signature image on a PDF	`pdf_sign_image`
Fill a PDF form	`pdf_form_fields` (see the field names first), then `pdf_form_fill`
Convert / resize / crop / rotate images	`image_convert`, `image_resize`, `image_crop`, `image_rotate`
Shrink images	`image_compress`
Watermark a photo	`image_watermark`
See or remove photo metadata (GPS!)	`image_exif_read`, `image_exif_strip`
Favicons, placeholders, palettes, ASCII art	`image_favicon_set`, `image_placeholder`, `image_palette`, `image_ascii_art`, `image_ascii_banner`
SVG → PNG, images ⇄ base64/data URIs	`image_svg_to_png`, `image_to_base64`, `image_from_base64`
Count words, change case, slugs	`text_word_count`, `text_case`, `text_slugify`
Clean lines, compare texts	`text_clean`, `text_diff`
Arabic: tashkeel, digits, Arabizi, wrong keyboard	`text_arabic`, `text_layout_fix`
Numbers as words (EN/AR), readability, citations	`text_number_to_words`, `text_readability`, `text_citation`
Escape strings for code, fancy Unicode, Morse, lorem ipsum	`text_escape`, `text_fancy`, `text_morse`, `text_lorem`
Post/SMS length limits	`text_social_count`
JSON / CSV / YAML / XML / SQL / HTML / CSS	`json_format`, `json_query`, `csv_to_json`, `json_to_csv`, `yaml_to_json`, `json_to_yaml`, `xml_format`, `sql_format`, `html_format`, `css_format`
XPath over XML, Markdown tables	`xml_query`, `markdown_table`, `markdown_table_parse`
Types from JSON, curl → code	`json_to_types`, `curl_to_code`
Hashes, Base64, URLs, JWT	`hash_compute`, `base64_encode`, `base64_decode`, `url_encode`, `url_decode`, `url_parse`, `jwt_decode`
IDs, regex, cron, timestamps	`id_generate`, `id_inspect`, `regex_test`, `cron_explain`, `time_convert`
Number bases, bytes, CSS units	`base_convert`, `bytes_convert`, `css_units`, `css_clamp`
Web dev bits	`meta_tags_generate`, `useragent_parse`, `gitignore_generate`
Networks and permissions	`subnet_calc`, `ip_in_range`, `chmod_calc`, `http_status`
Money maths	`calc_loan`, `calc_compound_interest`, `calc_percentage`, `calc_discount`, `calc_salary`, `calc_zakat`, `calc_split_bill`, `calc_cash_count`, `calc_fuel_cost`
Work, study, health	`calc_hours`, `calc_gpa`, `calc_bmi`, `calc_sleep`
Dates	`calc_age`, `calc_date_diff`, `calc_hijri`, `calc_due_date`
Other maths	`calc_unit_convert`, `calc_fraction`, `calc_stats`, `calc_equation`, `calc_primes`, `calc_roman`, `calc_expression`
Prayer times, qibla, world clock	`prayer_times`, `qibla_direction`, `time_world_clock`
Random picks, dice, teams, Secret Santa	`random_pick`, `random_dice`, `random_teams`, `secret_santa_draw`
Passwords	`password_generate`, `password_check`
Encrypt / decrypt (compatible with the website)	`crypto_text_encrypt`, `crypto_text_decrypt`, `crypto_file_encrypt`, `crypto_file_decrypt`
Hide a message in an image	`crypto_stego_hide`, `crypto_stego_reveal`
QR codes and barcodes	`qr_generate`, `qr_wifi`, `qr_read`, `barcode_generate`
Documents as PDF	`doc_invoice`, `doc_receipt`, `doc_business_cards`, `doc_labels`, `doc_calendar`, `doc_resume`, `doc_attendance_sheet`
Read a web page (uses the network)	`ai_web_fetch` (clean markdown; respects robots.txt; no JavaScript-only pages)
Any document → markdown you can read	`ai_html_to_markdown`, `ai_pdf_to_markdown`, `ai_docx_to_markdown`, `ai_xlsx_to_markdown`, `ai_pptx_to_markdown`
Text in images or scanned PDFs (English, Arabic)	`ai_ocr` (offline)
Fit text in a context window, split it for search/RAG	`ai_token_count`, `ai_chunk_text`
Near-duplicates, language, emails/phones/dates/money in text	`ai_text_similarity`, `ai_detect_language`, `ai_extract_entities`
Questions over CSV / Excel / JSON data	`ai_sql_query` (read-only SELECT across several tables), `ai_csv_profile`
JSON Schemas	`ai_json_schema_infer` (from samples), `ai_json_schema_validate`
Make Word, Excel, PowerPoint, PDF from content	`ai_markdown_to_docx`, `ai_create_xlsx`, `ai_create_pptx`, `ai_markdown_to_pdf` (Latin script only)
Charts and diagrams	`ai_chart` (simple spec or Vega-Lite → SVG/PNG), `ai_diagram` (`A -> B: label` text → SVG/PNG)
Zip, what is this file, compare images, audio/video info	`ai_zip`, `ai_file_identify`, `ai_image_diff`, `ai_media_info`

Do the arithmetic with the calculator tools instead of in your head: they use the same rounding and conventions as the website the user may compare with. Random tools (random_pick, random_dice, random_teams, secret_santa_draw) use secure randomness; pass a seed only when the user wants a reproducible draw.

Chaining

Pass one result's path straight into the next tool:

Scans to a small PDF: pdf_from_images → pdf_compress → pdf_page_numbers.
Share photos safely: image_exif_strip → image_resize (e.g. 1600 px) → image_compress.
Pull a table out of a report: pdf_to_text (pages) → reshape → json_to_csv.
Sign and send: pdf_form_fields → pdf_form_fill → pdf_sign_image → pdf_compress.
Wi-Fi poster: qr_wifi (PDF card) or qr_generate → pdf_from_images.
Scanned contract to editable Word: ai_ocr (write .txt) → tidy into markdown → ai_markdown_to_docx.
Spreadsheet to a report: ai_csv_profile → ai_sql_query → ai_chart (PNG) → ai_create_pptx with the chart image.
Long document Q&A: ai_pdf_to_markdown → ai_token_count → ai_chunk_text if it doesn't fit.

Write intermediate files to a temp folder and only the final file where the user asked.

Privacy promise

Tell users plainly when it matters: the files stay on their computer, nothing is uploaded, there is no account and no tracking. Website-port tools never touch the network. Tools marked openWorldHint (for example fetching a web page) do use the network; say so before calling them.

When to send a person to the web tool instead

Every tool's description ends with its web version (https://s3eidat.github.io/<slug>/). Point the user there when the job is visual or hands-on: drawing a signature, choosing a crop by eye, arranging pages by dragging, designing a QR code's look, live previews, scanning a QR with a camera, editing a CV layout, or when they have no AI client with this server. The web tools are free and also run entirely in the browser.

Limits to know

Document makers (doc_*) currently write Latin-script text only; for Arabic invoices/CVs send the user to the web tool.
Very large files: the server rejects inputs above its --max-mb (default 200 MB). Split work into parts.
pdf_compress works best on scanned/image-heavy PDFs; text-only PDFs are often already small.
Encrypted PDFs must be unlocked first (pass password where a tool offers it).