Using IconVectors with VS Code Copilot

Editing icons in IconVectors while you write code means switching windows and re-explaining context every time. Register the IconVectors MCP helper in VS Code and Copilot Chat, in Agent mode, can inspect the drawing that is open, create and edit geometry, render previews and export assets without you leaving the editor. In 2.0+ it can also browse the Icon Explorer and run the batch tools.

This page is about Copilot Chat in local VS Code on IconVectors 2.0+ for Windows, macOS, and Linux. GitHub Copilot CLI is a separate client with its own configuration and its own guide. On Linux the helper is at the path shown below, and the configuration works there too.

1. Start IconVectors

Windows (2.0+)

& 'C:\Program Files\Axialis\IconVectors\IconVectors.exe'

macOS (2.0+)

open -a /Applications/IconVectors.app

Linux (2.0+)

/usr/bin/IconVectors

Leave it running. VS Code launches IconVectorsMcp.exe itself, and the helper reaches the application on 127.0.0.1:61337 unless you changed Options/McpPort. One thing to check before you start: an organization can restrict MCP through Copilot Business or Enterprise policy, and no local setting overrides that.

2. Create the MCP configuration

Put the server in a workspace file, .vscode/mcp.json, or in your user configuration through MCP: Open User Configuration or MCP: Add Server from the Command Palette.

Windows example

{
  "servers": {
    "iconvectors": {
      "type": "stdio",
      "command": "C:\\Program Files\\Axialis\\IconVectors\\IconVectorsMcp.exe",
      "args": ["--port", "61337"]
    }
  }
}

macOS example

{
  "servers": {
    "iconvectors": {
      "type": "stdio",
      "command": "/Applications/IconVectors.app/Contents/MacOS/IconVectorsMcp",
      "args": ["--port", "61337"]
    }
  }
}

Linux example

{
  "servers": {
    "iconvectors": {
      "type": "stdio",
      "command": "/usr/bin/IconVectorsMcp",
      "args": ["--port", "61337"]
    }
  }
}

If you would rather not edit the file, the VS Code command line writes the same entry:

code --add-mcp "{\"name\":\"iconvectors\",\"type\":\"stdio\",\"command\":\"C:\\Program Files\\Axialis\\IconVectors\\IconVectorsMcp.exe\",\"args\":[\"--port\",\"61337\"]}"

Start the server from the controls in the configuration editor, or run MCP: List Servers and start it there. VS Code asks for workspace trust the first time a server starts and again after the configuration changes; starting it from its editor counts as an explicit trust decision, and MCP: Reset Trust lets you review it later.

3. Verify in Copilot Chat

Open Copilot Chat and switch it to Agent mode. Run MCP: List Servers to confirm iconvectors is started, then open the Chat tool picker to see its tools. Ask for the four discovery calls:

• Use the iconvectors MCP server to call app_ping and tell me whether the editor answered.
• Call app_getInfo and app_getCapabilities and summarize the running instance.
• Call app_get_workspace and tell me whether the Editor or the Explorer is active.

Before you allow a change, have Copilot inspect the state it is about to touch: document_getInfo and selection_get in the Editor, or the current folder with explorer_get_state in the Explorer.

4. Pick a tool set under the 128-tool limit

VS Code and Copilot are naming-compatible with all 166 production tools, but VS Code allows at most 128 enabled tools per chat request. IconVectors has no server-side profile to shrink the catalog, and a VS Code tool set groups tools for prompts without reducing what the server advertises, so the Chat tool picker is where you decide. Keep the discovery tools (app_ping, app_getInfo, app_getCapabilities, app_get_workspace) and the inspection tools for the state you are editing, then enable the family the job needs: Editor, Path, Paint and Stroke, Export, Explorer, or Batch. If more than 128 are enabled, deselect unrelated families before you send the prompt.

Since IconVectors 2.10 the catalog is 166 tools, and the eleven new ones are the Explorer's batch commands: all 14 of them are on the MCP surface, each as a plan request followed by an apply request, with the same preview counts, diagnostics and write safeguards as the desktop dialogs. The same client can therefore add badge overlays to a set, make Hot, Disabled and Dark Theme versions, export PNG and ICO files, convert a folder to currentColor or optimize its SVGs, as long as you read the plan before you approve the apply. The sequence is in Drive Icon Explorer and Batches from MCP.

5. Install the instructions and the skill

Copy vscode-copilot-AGENTS.md.template to the project root as AGENTS.md, or merge its short policy into the instructions you already keep; .github/copilot-instructions.md can carry the same policy. Help -> Open MCP Files Folder in IconVectors opens the per-user MCP helper-files folder, the first place to look for the template and the skill.

Install the complete iconvectors-mcp skill, references directory included, at one of these project locations (VS Code also documents personal roots under your .copilot, .agents or .claude folder):

• .github/skills/iconvectors-mcp/SKILL.md
• .agents/skills/iconvectors-mcp/SKILL.md
• .claude/skills/iconvectors-mcp/SKILL.md

The workflow and the schemas stay in the skill; the instruction files stay short.

6. Make a first bounded edit

1. Select the discovery tools plus a small Editor family in the tool picker.
2. Have Copilot inspect document_getInfo, selection_get and the DOM or style state the change depends on.
3. Ask for one bounded change and have it inspect the result again.
4. Save or export only when you asked for it, and only to a native Windows destination.

Several related edits go in one history transaction with the exact txId. Save and export, document replacement and Explorer file operations are outside document Undo.

Trust, approval and Windows safety

Trust and tool approval are separate decisions. Review approvals with Chat: Manage Tool Approval. MCP server sandboxing is not available on Windows, so keep the prompts for document replacement, save and export, and Explorer disk or batch operations rather than approving whole families to save clicks.

Troubleshooting

The server is not discovered

Validate the JSON and the scope, run MCP: List Servers, and confirm Agent mode and the server status.

The executable or the bridge fails

Check the full Windows path, start IconVectors, match the port, and confirm the intended application, workspace and document with app_getInfo when several instances exist.

Trust or approval blocks a call

Reset server trust or review Chat tool approval. Do not globally approve the destructive families to get past it.

More than 128 tools are enabled

Deselect unrelated tools in the picker and choose a task-focused set. Creating a tool set alone does not lower the enabled count.

You need logs

Run MCP: List Servers, select IconVectors, and choose Show Output.

Wrong state, a bad path, or too much output

Wait for Explorer navigation to settle and reinspect. Use native absolute Windows paths with JSON escaping, and ask for bounded or minified results.

GitHub's cloud coding agent runs in a hosted environment whose localhost is not your desktop, so it cannot reach this integration; only local VS Code and Copilot CLI can launch the helper. For current configuration details, see the official VS Code MCP documentation.

If IconVectors is not installed yet, download the free trial and run the first bounded edit with an icon open.