Using IconVectors with VS Code / GitHub Copilot

Overview

Visual Studio Code with GitHub Copilot can use IconVectors 2.00 through a local stdio MCP server on Windows. VS Code launches IconVectorsMcp.exe; the sidecar forwards calls over loopback to the already-running desktop application.

This page covers Copilot Chat in local VS Code. GitHub Copilot CLI has a different configuration and is documented in Using IconVectors with GitHub Copilot CLI. See MCP Integration for shared IconVectors behavior.

Prerequisites

  • Install a current Visual Studio Code release with GitHub Copilot access.

  • Start IconVectors 2.00 normally.

  • Confirm IconVectorsMcp.exe exists beside IconVectors.exe.

  • The default bridge is 127.0.0.1:61337. Match a custom Options/McpPort value when necessary.

An organization may restrict MCP through Copilot Business or Enterprise policy. Local configuration cannot override such a policy.

Configure VS Code MCP

Create .vscode/mcp.json in the workspace:

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

The command is the normal installed-product example. Replace the path or port when needed. Use MCP: Open User Configuration or MCP: Add Server from the Command Palette for user scope.

Start the server from the MCP configuration editor controls or run MCP: List Servers. Open Copilot Chat in Agent mode.

Verify the connection

Use MCP: List Servers to confirm that iconvectors is started, then use the Chat tool picker to confirm that its tools are available. Ask Copilot to call:

  • app_ping

  • app_getInfo

  • app_getCapabilities

  • app_get_workspace

Inspect the current Editor document and selection or the current Explorer folder before allowing a mutation.

Install instructions and the IconVectors skill

Copy vscode-copilot-AGENTS.md.template to the project root as AGENTS.md, or merge its short policy into the existing instructions. VS Code’s preferred repository instruction file .github/copilot-instructions.md can carry the same concise policy.

Install the complete distributed iconvectors-mcp skill, including its references directory, in one of these project locations:

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

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

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

VS Code also documents corresponding personal skill roots under the user’s .copilot, .agents, or .claude folder. Keep the full workflow and schemas in the skill rather than duplicating them in instructions.

First workflow

Try a small inspected edit:

  1. Select the application/state discovery tools and the small Editor family needed for the task.

  2. Inspect document_getInfo, selection_get, and relevant DOM/style state.

  3. Make one bounded change and inspect it again.

  4. Save or export only when a native Windows destination was requested.

Use one history transaction for related active-document edits. Save/export, document replacement, and Explorer filesystem operations are not covered by document Undo.

Working with 155 tools

VS Code and Copilot are naming-compatible with all 155 production IconVectors tools; canonical underscore names need no client-specific rename. However, VS Code currently permits at most 128 enabled tools per chat request. Use the Chat tool picker to select application/state discovery plus the relevant Explorer, Editor, Path, Paint/Stroke, Export, or Batch families.

Tool sets can group tools for prompts and agents, but they do not reduce what the MCP server advertises. IconVectors 2.00 has no server-side static or dynamic profile feature. If more than 128 tools are enabled, deselect unrelated tools; always retain the discovery and verification tools needed by the task.

Trust, approval, and Windows safety

VS Code requests workspace trust for an MCP server the first time it starts and after configuration changes. Use MCP: Reset Trust to review the choice again. Starting a server explicitly from its editor is itself an explicit trust action.

Tool approval is separate. Review it with Chat: Manage Tool Approval. MCP server sandboxing is not available on Windows, so keep prompts for document replacement, save/export, and Explorer disk or batch operations.

Troubleshooting

Server is not discovered

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

Executable or bridge fails

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

Trust or approval blocks the call

Reset server trust or review Chat tool approval. Do not globally approve destructive families.

More than 128 tools are enabled

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

Logs are needed

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

Wrong state, path, or excessive output

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

Local VS Code, Copilot CLI, and the cloud agent

This setup is local VS Code integration. Copilot CLI uses its own MCP configuration; see Using IconVectors with GitHub Copilot CLI.

GitHub’s cloud coding agent runs in a hosted environment. Its localhost is not the user’s Windows desktop, so it cannot directly reach the local IconVectors stdio/loopback server.

For current configuration details, see the official VS Code MCP documentation.