PDF Editor by PDF net

Apply a single editing command to the document in the editor session. `command` is a discriminated union keyed by `type` — each variant declares only its own fields. Supported types: `set_paragraph_text`, `add_page`, `remove_page`, `move_page`, `duplicate_page`, `add_paragraph`, `remove_paragraph`, `set_paragraph_bounds`. Use `describe_editor_pages` first to discover page and paragraph IDs. Form fields are not editable through this tool; use `fill_form_fields`. The call returns once the command is queued and its target page or paragraph has been confirmed to exist — an error here means nothing was queued. Queued commands are applied in call order; a following `describe_editor_pages` blocks until they finish and then reports the applied edits, or returns `status: busy` if they are still running. One command per call; prefer editing paragraph by paragraph over many consecutive calls. The user can revert changes in the editor UI. Requires an `editorSessionId` from `open_pdf_document_editor`. Returns an error if the editor was opened without a documentId and the user has not yet uploaded a file.

apply_editor_command

Produce a compressed copy of an existing document as a new library document named "<original> (compressed)". The original is unchanged. Each call creates a new document; retries will produce duplicates.

compress_pdf_document

Convert an existing document to another supported file format and return a temporary download URL (10 min). Does not create a new document in the library. A ZIP wrapper is returned when the conversion produces multiple files (e.g. PDF→PNG). Each call produces a new conversion artifact; retries will produce duplicates.

convert_pdf_document

Duplicate a document under a new name. The "destination" parameter controls placement: "current" (same folder as source, default) or "root". Folders cannot be copied. Each call creates a new document; retries will produce duplicates.

copy_pdf_document

Create a new folder. Each call creates a new folder; retries will produce duplicates.

create_folder

Move a document or folder to trash. Deleting a folder cascades - its full subtree (nested folders and documents) is trashed too. Trashed items are recoverable from the pdf.net web UI.

delete_resource

Return page metadata, paragraph IDs, paragraph bounds (`x`, `y`, `width`, `height`), `textWrapWidth`, paragraph-level style (`textAlign`, `lineHeight`, `textIndent`), and full text content for the document open in the editor session. Bounds are the read-only axis-aligned bounding box in PDF user space (Y axis up; `x`/`y` is its bottom-left corner, the origin `set_paragraph_bounds` accepts). `textWrapWidth` is the length the lines wrap at, in user space along the text direction, or "auto" when the paragraph widens with its content; it is the width `set_paragraph_bounds` accepts. It equals the bounds `width` only while the paragraph is not rotated or skewed. Text content is markdown: `**bold**`, `*italic*`, `^sup^` / `~sub~` (also `<sup>` / `<sub>`), and `<span style="...">…</span>` for per-run overrides of font-family, font-size (px), color, letter-spacing (px), -webkit-text-stroke, and vertical-align (px). Unstyled runs inherit the paragraph's first-character baseline. Pages that carry form fields also return `formFields`: the current value of every field on the page, the constraints on writing it, and the nearby text that may label it. `fill_form_fields` documents that shape and is the only way to change those values. Optional `pageNumbers` filter; omit to get all pages. Requires an `editorSessionId` from `open_pdf_document_editor`. Call it immediately after opening; it waits for the document to load. If the editor was opened without a `documentId`, returns `status: "waiting_for_upload"` until the user uploads a file.

describe_editor_pages

Get a pre-authenticated download URL for a stored document. The URL is valid for 10 minutes and single-use. No extra headers required.

download_pdf_document

Fill form fields of the document in the editor session — the only way to set field values; `apply_editor_command` edits page text and cannot touch them. Address each field by the `fieldName` from `describe_editor_pages`, never by `widgetId`: a value belongs to the field, so widgets sharing a `fieldName` share one value — that is a radio group, its options told apart by `checkedValue`. Write back the shape reported in `value`: a string for `text` and single-select `choice`, an array for multi-select `choice`, a boolean for `checkbox`, the option `checkedValue` for `radio` (as `value`, or as `optionValue` when it reads like plain text), `null` to clear. Constraints come with each field: `options` — send an option `value`, though its `label` is accepted and resolved — unless a combo box with `allowsCustomValue`; `maxLength` in characters when not null; `isMultiline`, without which no line breaks; `format`, e.g. a date pattern; `isComb`, a grid of `maxLength` one-character cells where a separator costs a cell; `canClear: false`, a radio group that cannot be emptied. Read-only fields and push buttons take no value. Identify a field before filling it. `fieldName` is often opaque (`Text1`, `f1_07[0]`) and a PDF field carries no label of its own, so `labelCandidates` are snippets of nearby page text, each tagged by where it sits — `inside` (overlapping the field, usually a placeholder), `left`, `right`, `above`, `below` — and ranked by proximity, closest first. They are a guess, not the label. Whenever the best candidate is missing, generic ("Date", "No.") or at odds with `fieldName`, work the page out instead: the paragraphs around the field, the section heading above it, and the fields on either side and their bounds all say what the form is asking for. Do not fill a field you cannot identify. Partial by design: each field is checked, the ones that pass are written together as a single undoable step, and the answer gives every written field its resulting value and every rejected field the reason and the constraints it must satisfy. Correct and re-send only the rejected ones. Nothing is written, and an error returned, only when every field fails. Fill related fields in one call rather than one call per field. This waits for the write, so a following `describe_editor_pages` sees the new values. Then report back to the user what landed in which field, naming the label you matched each one to, and ask them to check it before they sign, submit or send the document: matching their data to an unlabelled field is inference that can be confidently wrong, and only they can confirm their own details. They can also revert any change in the editor UI. Requires an `editorSessionId` from `open_pdf_document_editor`. Call it immediately after opening; it waits for the document to load. If the editor was opened without a `documentId`, returns `status: "waiting_for_upload"` until the user uploads a file.

fill_form_fields

Semantic search across the user's documents. Returns matching pages grouped by document. Optionally scope to specific documents.

find_in_documents