Inkscape version and location, server version, and open documents.

Create a new blank document and make it current. The viewBox matches width/height, so coordinates are in unit. background (e.g. '#ffffff') adds a full-page rect with id 'background'.

Open an existing SVG (.svg or .svgz) and make it current. Returns its outline, and notes when the file was normalised for editing without changing how it renders: a page whose viewBox doesn't match its width/height (or uses %), fill/stroke set on the root (new elements would inherit them), ids given to elements that had none.

Save the document as Inkscape SVG (to path, or where it was opened/last saved).

Outline of the document: layers, groups and elements with ids, fill/stroke as rendered (style, attributes or inherited), text, use targets (href) and real visual bounding boxes [x, y, width, height] in user units (measured by Inkscape). Layers/groups with more than max_children children are summarised (counts per type, first/last ids, bbox). To list one of them, pass layer (layer name or group id) and a larger max_children. find: a flat list of matching elements instead of the tree, e.g. {"fill": "#99cc32"} (colours compared in any notation), {"type": "text", "text": "total"}, {"type": "use", "href": "Parking"}, {"type": "symbol"} (symbols in defs, with titles), {"id_prefix": "card-"}.

Add one or more elements in a single call. Returns the new ids. Element spec keys — common: type, id, label, layer (name; created if missing), parent (group id), transform, style (css string or dict). Style shorthands: fill, stroke, stroke_width, opacity, fill_opacity, stroke_opacity, stroke_dasharray, stroke_linecap, stroke_linejoin, font_size, font_family, font_weight, font_style, text_anchor. Geometry per type: rect: x, y, width, height, rx, ry, fit_to, fit_padding, fit; circle: cx, cy, r; ellipse: cx, cy, rx, ry; line: x1, y1, x2, y2, marker_start, marker_end; polyline: points, marker_start, marker_end; polygon: points; path: d, marker_start, marker_end; text: x, y, text, line_height, vertical_anchor, width, halo, halo_width; group: -; arrow: x1, y1, x2, y2, shaft_width, head_width, head_length; image: x, y, width, height, href, object_fit, embed; use: x, y, width, height, href. points = [[x,y],...]. text supports '\n' for multiple lines; font_size is in user units. Coordinates are in the parent's system: inside a transformed layer or group (e.g. after layout moved it, or a scaled plan group) they are offset/scaled with it. use: href "symbol_id" places a symbol (or any element) of this document; "library.svg#symbol_id" copies that symbol from a file first (list a library's symbols: document_open it, inspect find {"type": "symbol"}); width/height scale a symbol that has a viewBox. clip: an element id (its current shape) or [x, y, w, h] — the element is cut to it and the clip then moves with the element; null removes it. text halo: '#ffffff' outlines the glyphs behind the fill so text reads over lines (halo_width default 0.3 x font size; 'none' removes). place: {"below": id, "gap": 1.5} (or above / left_of / right_of; "align": start|center|end on the other axis) puts the element beside the MEASURED box of another (real glyph extents, any script) and keeps it there when that element changes; inside repeat, ids are template names; null frees it. Responses warn about overlaps among the touched elements: text on text, text across a shape's edge, text crossed by a line (a text with a halo may cross lines). rect fit_to: [ids] sizes the rect around them after wrapping (fit_padding: n | [v, h] | [t, r, b, l]; fit: both | height | width, e.g. height keeps a card's width); it re-fits when those elements are edited (update_elements {"id": rect} re-fits after moves; fit_to: null frees it). defaults is merged into every element (only keys valid for its type), e.g. {"font_size": 2.2, "fill": "#2a8a4a", "layer": "Labels"}. elements_path: a JSON file (a list of specs, or {"elements": [...], "defaults": {...}}) instead of elements — for generated geometry, so it never passes through the conversation.

Stamp a block of elements once per data row — timelines, card grids, tables, legends, map symbols — in one call. template: element specs as for add_elements, drawn for the FIRST row. "{key}" in any string is replaced from the row ("{year}"; a value that is exactly "{w}" keeps the row's number), in geometry and style alike ("fill": "{colour}"); "{n}" is the row number (1-based) and "{i}" the index, so rows can't use the keys n and i. Ids are local names: "card" becomes card-1, card-2, ... A "parent" may name another template element (e.g. a group holding a card and its texts); fit_to and clip may name template elements of the same row. Each row goes into a group - moved by n-1 steps: step [dx, dy], or a grid with columns (step = [column pitch, row pitch]), filled row by row or, with order "column", column by column. cell ["col", "row"]: each row is placed by its own 1-based column/row values (gaps and fractions allowed: periodic tables, calendars, timetables, lanes); the template is drawn for cell (1, 1) and step is the [column, row] pitch. Components (a character, a node, a symbol placed at data positions): a template group with "transform": "translate({x},{y}) scale({s})" and step [0, 0], parts drawn around a local origin, pose/shape parts as placeholders ("d": "{arms}"). mirror {"x": 148.5, "rows": "even"|"odd"|"all"} (or "y") mirrors those rows about the axis: shapes are reflected (pointers flip), texts and groups keep their reading direction and move as blocks — group a card with its texts so they cross together. Per element "mirror": "reflect"|"block"|"none" overrides. rows_path: a .json (list of row objects) or .csv file (header line = keys, numbers parsed) instead of rows, so data never passes through the conversation. Returns the row groups, ids per template name (runs shortened to "card-1..card-12"), wrapped_lines, fitted.

Place a file into the current document. An SVG (e.g. geometry a script generated) becomes one group scaled to this document's units, its top-left at at (default 0,0); width or height scales it (both: stretch). Its layers become labelled groups, its defs join ours, clashing ids get "-" in front. An image (png/jpg/gif/webp/bmp) becomes an image element: natural size at 96 dpi unless width/height (one keeps the ratio), linked by default (embed: true stores the pixels in the SVG), object_fit contain|cover|fill for a given box. Relative paths start at the document's folder. For whole documents use document_open; for data rows see repeat rows_path.

Divide a region into named cells — comic panels, dashboard tiles, poster columns, table grids — without computing any positions. region: "page", an element id (its measured box) or [x, y, w, h]; margin insets it (one number, [vertical, horizontal] or [top, right, bottom, left]). rows: one entry per row, either a number of equal columns or a list of column ratios, e.g. [[2, 1], [1, 1], [1, 2]]; heights: row ratios (default equal); gutter: space between cells (number or [horizontal, vertical]). Creates one rect per cell, -1, -2, ... row by row, in layer — invisible unless style gives e.g. {"fill": "#fff", "stroke": "#000", "stroke_width": 0.6, "rx": 2}. Use the ids as targets: clip ("clip": "cell-5"), align/layout ("to": "cell-2"), place, fit, connect. An element region is its measured box (stroke included). Tables: split the header strip into columns (e.g. rows [[3, 1, 1]]), then repeat the data rows with step [0, row pitch] and texts at the returned column edges (numbers: text_anchor end at the right edge). Returns the cells [x, y, w, h] and their ids per row.

Change existing elements. Each update is {"id": ..., }; only the given keys change. Style shorthands merge into the existing style. Set transform to "" to clear it.

Delete elements (and their children) by id. Connectors attached to them and their labels are deleted too; all removed ids are returned. Defs only they used (symbols of an imported library, gradients, clip paths) go too: defs_removed counts them.

Draw arrows/lines between elements that stay attached when things move (align/layout/ update_elements/page_fit). Default: native Inkscape connectors — clipped to the real shape (circles etc.) and still live in the Inkscape GUI. With from_side/to_side/via you control the route (e.g. a loop diagram: {"from": "condenser", "to": "valve", "from_side": "left", "to_side": "top", "routing": "elbow"}); those attach at the middle of the chosen side of the bounding box, as do connectors with start_gap/end_gap. Returns the connector ids, their layers, label ids and warnings.

Arrange items in a row, column or grid with a gap — no coordinate maths needed. An item is an id or a list of ids that move together, e.g. ["box1", "box1_label"], or {"ids": [...], "anchor": "fig1-border"} to arrange by that member's box (plot frames line up even when their tick labels differ in width); the rest moves along. gap is a number or [horizontal, vertical]. align places items on the cross axis (grid: within cells). The block stays where the first item is, or its top-left goes to at [x, y], or it is aligned to to ('page' or an element id) using horizontal/vertical/margin. Connectors follow.

Geometry operations performed by Inkscape on the given ids. For difference, the top-most (later in document order) object is subtracted from the bottom one. to_path converts shapes and text into paths. union/intersection/exclusion/difference keep the BOTTOM object's id, style and layer; combine keeps the TOP object's (E22). result lists the ids that hold the outcome.

Escape hatch: run raw Inkscape actions (e.g. "object-align:left last", "transform-rotate:30") after selecting select ids. File, export, window and quit actions are blocked. Extensions (org.inkscape.*) always act on the whole document, so they can't be combined with select.

Resize the page to fit the drawing (or only ids) plus margin (one number, [vertical, horizontal] or [top, right, bottom, left], user units). All content moves together so the page keeps its 0,0 top-left; full-page background rects are resized, connectors follow.

Set the page size in user units (e.g. 210 x 297 for A4 in a mm document). anchor='center' keeps the drawing centred on the new page; 'top-left'/'none' leave content where it is. Backgrounds are resized.

Draw a grid / graph paper inside rect [x, y, w, h] — linear or logarithmic per axis — without computing any line positions. Axis specs: linear: {"scale": "linear", "major": 10, "medium": 5, "minor": 1, "label_start": 0, "label_step": 1} (spacings in user units, each a whole multiple of the finest; labels on major lines) log: {"scale": "log", "cycles": 3, "subdivisions": "standard" | "fine" | "integers", "start": 10} (decades major, 2..9 medium, subdivisions minor; "start" = value at the origin (default 1); "labels": "decades" (start, start*10, ...; default when start is given) | "paper" (1..9 per cycle)) "reverse": true flips an axis (default x left→right, y bottom→top); "lines": false keeps the axis (labels, plot mapping) but draws none of its gridlines, e.g. vertical-only lines for a bar chart. weights: {"major", "medium", "minor"} stroke widths (defaults 0.45/0.22/0.08 mm); border: stroke width of the frame (default 0.6 mm, 0 = none). labels: {"sides": ["left", "bottom"], "font_size", "gap", "color", "font_family", "bold_major", "x_title", "y_title", "title_font_size"} — placed outside the grid, centred on their lines (measured); major labels are bold unless "bold_major": false; titles go below / left (rotated) of the tick labels. Result: one path per weight class in layers ' minor/medium/major', labels in ' labels'. Use plot with grid= to draw data on it.

Plot data on a grid made with the grid tool, in DATA values — no coordinate maths. grid: that grid's id_prefix. Each series: {"points": [[x, y], ...], "line": true, "stroke": "#1f77b4", "stroke_width", "stroke_dasharray", "marker": "circle"|"square"|"diamond"|"none", "marker_size", "marker_fill" (default white), "point_labels": ["", "COP 3.2", ...] (one per point, null/"" to skip), "label_offset": [dx, dy], "label_anchor": "start"|"middle"|"end", "label_halo": "#ffffff"|"none", "label_font_size", "label_color", "font_family", "id", "layer"}. Labels: label_offset is from the point to the label's anchor, and the label is vertically CENTRED on point + dy (dy = 0 → centred on the point). Anchor defaults to start for dx >= 0, else end. Without label_offset the label goes right of the point, on the side the line is not heading to. Labels carry a halo stroke (default white) so the line can cross them — set label_halo "none" (or the background colour) for labels on dark fills; recolouring a label later does not remove its halo (use stroke: "none"). Ids follow the series id: -line, -marker-, -label- (k = 1-based point index). Returns per series those ids and the points in user units (for annotations); warns about points outside the grid.

Change stacking order (what is drawn on top). front/back: top/bottom within the element's own layer or group. forward/backward: one step past the next object it overlaps (visible change). above/below: directly above/below target, moving into target's layer/group if needed while keeping the visual position. Several ids keep their relative order. Returns each id's position.

Move elements into a layer (by name; created on top if missing) or into a group/layer by id, at its top or bottom. They keep their relative order and stay visually where they were.

Align elements to the page, to each other, or inside another element — e.g. centre a label in a box: {"ids": ["label"], "to": "box", "horizontal": "center", "vertical": "middle"}. Operations run in order, each seeing the previous moves. Returns the moves [dx, dy] and new bboxes in user units.

Export via Inkscape. Format defaults to the file extension. What is exported: region [x, y, w, h] (user units, everything visible); or ids — only those objects cropped to them (only_ids=true, default) or the area around them with everything visible (only_ids=false); otherwise area 'page' or 'drawing'. For PNG set dpi or width/height (px); background e.g. '#ffffff' (default transparent).

Render to a PNG image you can look at (white background, longest side = max_size px). Zoom in with region [x, y, w, h] in user units, or with ids (the area around them, everything still visible). only_ids=true draws just those objects.