MCP Integration

Version 1.40 makes IconVectors available through the Model Context Protocol (MCP). The feature is aimed at local, developer-oriented automation: an MCP-capable AI client can connect to the running editor, inspect the current document, create geometry, edit SVG structure, drive icon-level commands, render previews, and export code or files without bypassing the application’s own document model.

Once the MCP server is configured in your AI agent, just type a prompt to work with icons. For example, the prompt below creates an info icon:

Using IconVectors MCP commands, create a new 32x32 document named "info icon", draw a blue circle
in the center, use a color from the "Axialis Colors" palette, draw a large info "i" white letter
using 2px round corners, center vertically and horizontally the "info" group, and export it as SVG.

Here is the result (Codex CLI on left, Axialis IconVectors on right):

Creating an Icon with Codex spacer

How MCP works in IconVectors

IconVectors uses two cooperating processes:

  • The desktop editor owns the document, UI state, undo history, rendering pipeline, and local bridge:

    • Windows: IconVectors.exe

    • macOS: IconVectors.app (main executable inside the bundle)

  • The companion MCP server is the local helper process launched by the AI host:

    • Windows: IconVectorsMcp.exe

    • macOS: /Applications/IconVectors.app/Contents/MacOS/IconVectorsMcp (or the equivalent path inside your app bundle)

  • A host such as Codex, Claude, VS Code, or Cursor launches the helper as a local stdio process.

  • The server forwards MCP tool calls to the running editor over a local TCP connection. The default port is 61337.

  • Responses come back as structured data, so the AI client can inspect state, react to errors, and chain operations safely.

Because the MCP layer is backed by the editor itself, the AI sees the same document state as the user. It is not editing detached SVG text in memory unless you explicitly replace the document through a raw SVG command.

Typical workflow

  1. Start IconVectors on your platform.

  2. Configure your MCP host to launch the local MCP helper with the same bridge port used by the editor:

    • Windows: IconVectorsMcp.exe --port 61337

    • macOS: /Applications/IconVectors.app/Contents/MacOS/IconVectorsMcp --port 61337

  3. Let the host discover the available tools.

  4. Use those tools to inspect or mutate the open document.

  5. Undo or redo MCP-driven edits from IconVectors itself, because many write operations use the same command system as the UI.

What MCP is useful for

MCP is especially useful when you want an AI client to work with the real editor instead of generating SVG blindly. Common scenarios include:

  • create, draw, and edit SVG vector icons from natural-language instructions,

  • inspect the SVG DOM, selection, styles, gradients, and editable path data before changing anything,

  • automate palette operations, recoloring, badge placement, alignment, and other repetitive tasks,

  • render preview bitmaps for review or downstream tooling,

  • export the current drawing as SVG, Minified SVG, SVG Symbol, XAML, WPF Geometry, VectorDrawable, SwiftUI, VueTS, ReactTS, C++, or bitmap outputs.

Supported host pattern

The current design is a local stdio MCP server. It works well with local MCP-compatible desktop or CLI hosts such as:

  • Codex and other OpenAI tooling that can launch a local MCP server on the same machine,

  • Visual Studio Code with GitHub Copilot,

  • Cursor,

  • Claude Desktop,

  • Claude Code,

  • other MCP hosts that can start a local executable on the same machine and speak stdio MCP.

Hosted web/API connectors are different:

  • ChatGPT connectors and the OpenAI Responses API expect an HTTP-accessible MCP endpoint.

  • Anthropic’s hosted API connector also expects a remote HTTP MCP server.

  • To use IconVectors there, place an HTTP wrapper or relay service in front of the local bridge.

Installed MCP helper files

If your IconVectors installation includes the optional MCP helper files, the cleanest layout is:

  • Windows binaries in C:\Program Files\Axialis\IconVectors

  • Windows helper files in %APPDATA%\Axialis\IconVectors\MCP

  • macOS application bundle in /Applications/IconVectors.app

  • macOS helper files in ~/Library/Application Support/Axialis/IconVectors/MCP

On a typical Windows system, %APPDATA%\Axialis\IconVectors\MCP expands to something like C:\Users\[username]\AppData\Roaming\Axialis\IconVectors\MCP.

You can open that folder directly from IconVectors with Help ‣ Open MCP Files Folder.

Recommended setup-guide files:

  • codex-implementation-instructions.md

  • claude-implementation-instructions.md

  • vscode-copilot-implementation-instructions.md

  • cursor-implementation-instructions.md

Recommended companion files:

  • iconvectors-mcp-command-reference.md

  • codex-AGENTS.md.template

  • iconvectors-codex-skill\SKILL.md

  • claude-CLAUDE.md.template

  • vscode-copilot-AGENTS.md.template

  • cursor-agent-instructions.template.md

The setup guides tell the user how to configure each AI host. The companion files are optional context files the user can copy into a repository or paste into agent instructions to help the host choose IconVectors MCP tools more reliably.

Useful files to give the AI agent

For most end users, the most helpful files are:

  • iconvectors-mcp-command-reference.md - human-readable list of all MCP commands

  • the host-specific setup guide - tells the user where to put the MCP server command/configuration

  • a host-specific instruction template such as codex-AGENTS.md.template or claude-CLAUDE.md.template - helps the AI prefer IconVectors MCP when icon work is involved

  • for Codex specifically, the optional iconvectors-codex-skill\SKILL.md file - provides a stronger reusable IconVectors workflow with defaults for document creation, palette usage, pixel-perfect drawing, and SVG-first output

iconvectors-mcp-tools.json is also useful, but mainly for advanced users or debugging. It contains the exact JSON schemas for every MCP tool and normally lives beside the other helper files in the per-user MCP folder.

Codex can also benefit from an optional SKILL.md skill pack. For the other hosts, native instruction files such as AGENTS.md or CLAUDE.md remain the better fit for real customer setups.

Available MCP commands

The 1.40 MCP specification exposes 137 commands. They are grouped below by family.

  • ``action`` - generic application commands: action.list, action.invoke

  • ``app`` - health checks, runtime information, and capability discovery: app.ping, app.getInfo, app.getCapabilities

  • ``code`` - generated source code retrieval: code.get

  • ``create`` - new SVG primitives and text-as-paths creation: create.group, create.rect, create.ellipse, create.line, create.path, create.polyline, create.polygon, create.textAsPaths

  • ``document`` - document lifecycle, SVG access, and file placement: document.new, document.open, document.close, document.save, document.saveAs, document.getInfo, document.getSvg, document.setSvg, document.placeFiles, document.placeTraceBitmap

  • ``dom`` - raw SVG DOM inspection and mutation: dom.getElement, dom.getSubtree, dom.findById, dom.query, dom.getDefs, dom.createElement, dom.deleteElement, dom.setAttributes, dom.setStyle, dom.reparent, dom.reorder

  • ``element`` - transforms, grouping, naming, visibility, and deletion: element.move, element.scale, element.rotate, element.reflect, element.center, element.centerHorizontally, element.centerVertically, element.alignHorizontalLeft, element.alignHorizontalCenter, element.alignVerticalCenter, element.distributeHorizontally, element.matrix, element.group, element.ungroup, element.duplicate, element.setVisible, element.rename, element.setName, element.getName, element.delete

  • ``export`` - file export commands: export, export.toBitmap

  • ``gradient`` - gradient creation, inspection, and application: gradient.create, gradient.addStop, gradient.updateStop, gradient.removeStop, gradient.setAngle, gradient.apply, gradient.list, gradient.get, gradient.createLinear, gradient.createRadial, gradient.applyToElement

  • ``history`` - undo/redo state and explicit transactions: history.getState, history.undo, history.redo, history.beginTransaction, history.commitTransaction, history.rollbackTransaction

  • ``icon`` - whole-icon resizing, color operations, variants, badges, and palette-aware recoloring: icon.resize, icon.resizeCanvas, icon.colors.brightnessContrast, icon.colors.hueSaturation, icon.colors.levels, icon.colors.desaturate, icon.colors.invert, icon.addAllColorsToPalette, icon.recolorWithPalette, icon.variants.makeDarkThemeVersion, icon.variants.makeDisabledVersion, icon.variants.makeHotVersion, icon.badge.add, icon.badge.cutout, icon.badge.shadow, icon.badge.placeOverlay, icon.badge.makeCutout, icon.badge.addShadow

  • ``layout`` - centering, alignment, and distribution helpers: layout.center, layout.align, layout.distribute

  • ``palette`` - palette inspection and editing: palette.list, palette.getActive, palette.setActive, palette.import, palette.export, palette.addColor, palette.removeColor, palette.addAllIconColors

  • ``path`` - path conversion and low-level anchor editing: path.convertToPath, path.outlineStroke, path.offset, path.simplify, path.makeCompound, path.releaseCompound, path.getEditable, path.setEditable, path.moveAnchor, path.moveHandle, path.deleteAnchor, path.setAnchorType, path.roundCorner

  • ``prefs`` - preference read/write operations: prefs.get, prefs.set

  • ``render`` - preview rendering: render.previewPng

  • ``selection`` - selection inspection and control: selection.get, selection.set, selection.clear, selection.selectAll

  • ``style`` - fill, stroke, opacity, and stroke parameter updates: style.get, style.setFill, style.setStroke, style.setOpacity, style.setStrokeWidth, style.setLineCap, style.setLineJoin

  • ``view`` - canvas and panel visibility helpers: view.centerCanvas, view.showGrid, view.showPreviewPanel, view.showLayersPanel, view.showColorPalettesPanel, view.setGridVisible, view.setSnapToGrid

See also