Is Inkscape (Claude) running; which documents are open; what is frontmost; what the user has selected.

Start Inkscape (Claude) if needed and open path in it (an .svg; Inkscape also opens .pdf/.ai/.eps).

The object tree with ids, labels, text snippets and live bounding boxes (mm).

root: start below this id (default: the whole page). depth: levels to expand; deeper content is summarised as counts. refresh: force re-reading the document (normally automatic).

Find objects by any combination of: text content (case-insensitive substring), inkscape label, id regex, tag (text, g, rect, image, path…), a style substring (e.g. 'fill:#4b80ad'), or a page region [x0, y0, x1, y1] in mm (within = fully inside, overlaps = touches). top_level_only drops matches nested inside another match.

Everything about specific objects: tag, label, where it sits in the tree, bounding box, every attribute (embedded images summarised), parsed style, full text, and its children. xml=True adds the raw XML (truncated to max_chars).

Read what the user currently has selected (ids + boxes), or replace the selection with set_ids (useful to show the user which objects you mean).

Render the LIVE document to an image you can see.

region_mm [x0, y0, x1, y1] or around=[ids] (+margin) choose the area; default is the whole page. only=True hides everything except the around objects. annotate draws each object's box and id on the image: pass a list of ids, 'children:' for that object's children, or 'top' for the main parts in view (the shallowest level showing several objects, wrappers skipped) — the fastest way to learn which id is which.

Screenshot of the Inkscape window itself (the user's view: zoom, selection handles, any open dialog). Use when something seems stuck or to see what the user is looking at.

What the user changed since you last looked or edited (or since since_revision): objects added, removed, moved or resized (exact, cheap). Reports the current revision and makes it the new baseline. deep=True also re-reads the document to report text and style edits (slow on posters).

Align objects exactly. edges: 'left' | 'hcenter' | 'right' | 'top' | 'vcenter' | 'bottom', or one of each ('left top'). to: an object id to align to (it stays put), or first | last | biggest | smallest | page | drawing | selection. to_mm: align that edge to a page coordinate instead. as_group moves the ids as one block. expected_revision: refuse if the document changed since.

Space objects along x or y. Without gap_mm: Inkscape's equal distribution between the outermost objects (mode gap | centers | left | right | top | bottom). With gap_mm: stack them in their current order with exactly that gap, starting at the first object (or at start_mm).

Move by (dx_mm, dy_mm) and/or place so the anchor point ('top-left', 'center', 'bottom-right', 'top-center'…) of the ids' joint box lands at (x_mm, y_mm). each=True places every object separately.

Resize to an exact visual width or height (or by a scale factor), keeping the anchor corner fixed. keep_aspect=False with both width_mm and height_mm stretches (goes through the slower extension).

Set CSS style properties on objects, e.g. {"fill": "#4B80AD", "font-size": "11pt", "stroke-width": "0.5", "opacity": "0.8"}. A null value removes the property. Applies to groups too (children inherit unless they override — inspect them if a colour does not take).

Set (or with null, remove) XML attributes: 'inkscape:label', 'width', 'rx', 'xlink:href', 'id'…

Change text in ONE undoable batch. edits: [{"id": text id, "text": "new text; \n for new lines"}] keeps the first run's style. replace: [{"old": "…", "new": "…", "ids": optional scope}] swaps a substring inside existing runs, preserving mixed formatting (bold words, superscripts).

Place a figure file into the document with its top-left at (x_mm, y_mm), scaled to width_mm or height_mm (aspect kept). .svg is inlined as editable vector (ids namespaced so nothing collides); .png/.jpg are embedded (embed=False links the file instead).

Draw anything by writing SVG markup: shapes, lines, arrows, text, groups, gradients. With x_mm/y_mm the markup's coordinates are units ('mm' or 'px') from that page point, so '' at x_mm=20 is a 40×10 mm bar at 20 mm. Without them the markup is written as-is in the parent's own user units. Colliding ids are renamed.

Rearrange objects. op: delete | duplicate | group | ungroup | pop_out (out of its group) | raise | lower | top | bottom (z-order) | clone | unclone | reparent (move into parent at index, keeping the visual position). group takes optional new_id and label.

Undo or redo n steps in the live document (the same stack as the user's Cmd+Z). Pass expected_revision so an undo never takes back something the user did after your edit.

Run any of Inkscape's ~1,070 actions, or search them. run: ["select-all", ["object-align", "left page"], ["transform-rotate", 90]] — returns what Inkscape printed, any complaint it logged (an action it rejected makes this an error), and whether the geometry changed (Inkscape ignores some malformed arguments silently). search: substring over action names and descriptions. Most act on the current selection; use selection(set_ids=…) first. To write files, prefer export.

Run Python (inkex) against the live document in one undoable step — anything the other tools cannot express. In scope: svg (root element), by_id(id), selected (ids), inkex, etree, Transform, frames (frames.content(el) maps an element's coordinates to page px), to_px('3mm'), qname('inkscape:label'), unique_id(base), parse_style/style_str. print() output and result = … come back. readonly=True skips the document reload (fast inspection). timeout_s: give up waiting after this long (the result is then UNCERTAIN COMPLETION: the code may still finish and apply).

Export the live document (unsaved changes included) to PDF, SVG, EPS, PS or PNG; the format comes from the file suffix (or format=). What to export: area='page' (default) or 'drawing' (everything drawn); or ids=[…] for those objects (only=False keeps whatever overlaps them); or region_mm=[x0, y0, x1, y1] (PNG only). dpi: PNG resolution, and the resolution filter effects are rasterised at in PDF/EPS/PS. text_to_path=True turns text into outlines (for printers that cannot take fonts). background: PNG background colour (default transparent). The file is read back afterwards: page size in mm, pixel size, fonts embedded.

Write the live document (unsaved changes included) to path as Inkscape SVG. Does not change which file the window is editing; the user saves the real file with Cmd+S.