scholarocr
/scholarocr
Odczytuje, wyciąga i weryfikuje treść zeskanowanych dokumentów akademickich, np. zdjęć stron, egzaminów i wzorów matematycznych, przez ScholarOCR.
- Autor
- adiroy911 ↗
- Repozytorium
- scholarocr ↗
- Licencja
- MIT
- Ostatnia zmiana repozytorium
- 04.11.2026
Ocena i bezpieczeństwo
- Pliki w paczce
- 5
- Skrypty
- 0
- Komentarze HTML
- 1
- Adresy URL w SKILL.md
- 0
Brak sygnałów. Repozytorium zmienione 04.11.2026, bez odwołań do wycofanych modeli i API.
MIT, otwarta: można używać i zmieniać, z zachowaniem informacji o autorze.
- SKILL.md zawiera komentarze HTML (1). Nie widać ich w podglądzie Markdown, a model je czyta.
Co robi i kiedy
Odczytuje, wyciąga i weryfikuje treść zeskanowanych dokumentów akademickich, np. zdjęć stron, egzaminów i wzorów matematycznych, przez ScholarOCR.
Oryginalny opis (EN)
Read, extract and verify content from scanned or complex academic documents (scanned PDFs, photos of pages, handwritten notes, exam/question papers, textbooks, research papers, math/chemistry/physics material) through the ScholarOCR MCP server or CLI. Use whenever a task depends on the contents of such a document, when a PDF has no reliable text layer, when equations/tables/diagrams must be read accurately, or when the user wants a scanned document converted to Markdown, searchable PDF, DOCX, LaTeX or spreadsheets.
Jak działa
- When to use it
- Core workflow (MCP tools: mcpscholarocr)
- Rules
- CLI fallback (no MCP available)
- References (read when needed)
Podgląd SKILL.md
---
name: scholarocr
description: Read, extract and verify content from scanned or complex academic documents (scanned PDFs, photos of pages, handwritten notes, exam/question papers, textbooks, research papers, math/chemistry/physics material) through the ScholarOCR MCP server or CLI. Use whenever a task depends on the contents of such a document, when a PDF has no reliable text layer, when equations/tables/diagrams must be read accurately, or when the user wants a scanned document converted to Markdown, searchable PDF, DOCX, LaTeX or spreadsheets.
---
# ScholarOCR
ScholarOCR is a shared OCR/intelligence backend. It turns academic documents into a structured,
source-traceable representation (pages → elements with type, text, LaTeX, tables, bounding boxes,
uncertainty). **Always go through the engine and never guess unreadable content.** It needs no API
key: in host mode (the default without a key) it does all deterministic work and hands you, one page at
a time, the page images that need vision; you transcribe them into its schema and it validates, stores
and verifies the result.
## When to use it
Use it when the answer depends on what a document says and any of these hold: the file is a scan
or photo; it contains equations, chemical formulas, tables, diagrams or handwriting; the user asks
for an editable/searchable version; or you would otherwise read page images yourself. For a plain
born-digital PDF of ordinary prose, ScholarOCR is still cheap (no model calls for pages with a
reliable text layer) and gives you page/element references.
## Core workflow (MCP tools: `mcp__scholarocr__*`)
1. **Reuse first**: `list_documents(query=<filename>)`. If it is already processed, skip to step 3.
Calling `ocr_document` again is safe — unchanged pages come from cache — but costs a round trip.
2. **Extract**: `ocr_document(file_path=..., document_type=?, context_hint=?, pages=?)`.
- Pass hints you actually know ("Class 11 chemistry exam", `document_type="handwritten_notes"`).
- Large documents return a `job_id`; poll `get_document_status(job_id=...)` every so often
(don't spin). Process only the pages you need (`pages="40-55"`) when the task is local.
- The result says `mode`. With `"host"`, pages in `awaiting_transcription` need YOUR vision:
loop `get_page_image(document_id)` (one image + rules + element shape) -> transcribe exactly ->
`submit_page_content(document_id, page_number, elements=[...])`. Boxes are
`box_2d=[ymin,xmin,ymax,xmax]` in 0-1000 of that image; unreadable parts are type `uncertain`
with uncertainty `unresolved`. A rejected submission lists every problem (path, problem, fix):
fix all and resubmit the whole page. Follow `next_waiting_page`; never re-request a done page.
Optional second pass: `get_source_region(element_id)` then `submit_review(...)` for
`review_suggested` elements.
3. **Retrieve narrowly**: `get_document_content(document_id, pages=..., element_types=[...])`
(markdown by default, paginated via `next_cursor`). Use `get_page_content` for the full IR of a
page (bounding boxes, uncertainty reasons, quality, transforms).
4. **Check uncertain content against the source** before relying on it: anchors look like
`<!-- el:ID uncertainty=high -->`; `⟦unclear: …⟧` marks unreadable regions. Call
`get_source_region(document_id, element_id=ID)` to see the original crop and read it yourself.
5. **Verify** with `verify_document(document_id)` before high-stakes use (solutions, conversions,
grading). Report ERROR/WARNING findings that affect the user's task.
6. **Export only on request**: `export_document(document_id, format)`; no OCR is re-run.
7. **Cite pages**: refer to content as "page N" (and element ids when useful) so the user can check.
## Rules
- Uncertainty labels (low/medium/high/unresolved) are ordinal, **not probabilities**. Never present
medium/high/unresolved content as certain; say what is uncertain and show the source when it matters.
- Never fill in `unresolved` text. Tell the user it was unreadable and where (page, element).
- Don't delete anything unless the user explicitly asks. `delete_document(mode="all", confirm=true)`
removes ScholarOCR's preserved copy; the user's original file is never touched.
- With a key, `verify_document(review=true)` and `ocr_document(detail="high")` spend extra model
calls. In host mode `verify_document(review=true)` returns `review_candidates` for you to re-check.
- Host-mode transcription is a single pass by you, a general model; the key-mode pipeline is a
dedicated OCR prompt plus a second model. Say so when exactness matters.
- `classify_document` needs a key; in host mode classify from the content yourself.
- File access is limited to the server's authorized directories; on `access_denied`, tell the user
to append the folder to `SCHOLAROCR_ALLOWED_DIRS` in the settings file the error names (keeping the
existing entries) rather than copying files around.
## CLI fallback (no MCP available)
`scholarocr --json process FILE [--pages 1-10] [--type exam-paper] [--context "..."]`, then
`scholarocr content DOC_ID [--pages ..] [--types equation,table]`, `scholarocr region DOC_ID
--element ID --out crop.png` (then view the PNG), `scholarocr verify DOC_ID`,
`scholarocr export DOC_ID --format docx`. Exit code 6 = no model key (pages await transcription,
which only an MCP host or a key can do) or model failure, 7 = finished with problems.
## References (read when needed)
- `references/mcp-tools.md` — every tool, parameters and return shapes.
- `references/workflows.md` — recipes: solve a textbook exercise, digitize question papers,
handwritten notes, research papers, batches, large documents.
- `references/output-formats.md` — choosing and configuring export formats.
- `references/verification-and-errors.md` — verification findings, uncertainty rules, source
traceability, error codes and recovery.
Pliki w paczce (5)
- SKILL.md 6 KB
- references/mcp-tools.md 6 KB
- references/output-formats.md 2 KB
- references/verification-and-errors.md 3 KB
- references/workflows.md 3 KB
Wersje
W bazie nie ma innych skilli o nazwie scholarocr.
Migawka bazy z 13.11.2026 · wersja paczki 0e21ce1ab541 (skrót SHA-256 całej paczki)
Pobrania w ostatnich 30 dniach: 0 · ranking
Zgłoś problem z tym skillem: szkodliwa treść, błędny opis albo prośba autora o usunięcie.
Podobne skille
- 86ocr
Wyciąga tekst ze skanu PDF, zdjęć stron, notatek odręcznych, arkuszy egzaminacyjnych i podręczników przez ScholarOCR. - 96ocr-export
Eksportuje dokument już przetworzony przez ScholarOCR do JSON, Markdown, PDF z tekstem, DOCX, LaTeX, CSV lub XLSX. - 100pdf-extraction
Wyciąga tekst z PDF-ów przez zdalne API OCR Asta, także ze skanów dokumentów. - 100pdf-ocr
Wyciąga tekst ze zeskanowanych plików PDF za pomocą rozpoznawania znaków (OCR), w wielu językach i także z niskiej jakości skanów. - 100mineru-pdf-to-markdown
Konwertuje PDF-y i dokumenty naukowe do Markdowna przez MinerU przed czytaniem i analizą, z zachowaniem równań, tabel i zeskanowanych stron, których zwykła ekstrakcja tekstu nie oddaje. - 100handwritten-notes-to-latex
Zamienia odręczne notatki z wykładu w PDF na złożony dokument LaTeX zgodny z oryginałem: numeracja, tytuły, twierdzenia i dowody.