ChatGPT desktop supports local MCP servers over STDIO.

1. Open the ChatGPT desktop app.
2. Open Settings.
3. Select Plugins, then MCPs.
4. Click Add. If prompted for a server type, choose STDIO.
5. Enter:
   • Name: ifc-mcp
   • Command to launch: npx
   • First argument: -y
   • Second argument: ifc-mcp
6. Leave the environment variable fields empty and save the server.
7. Restart the ChatGPT desktop app.
8. Start a new chat, type /mcp, and confirm that ifc-mcp is connected.

For more detail, see OpenAI's MCP documentation.

Option 1: Use the Claude Desktop UI

1. Open Claude Desktop.
2. Open Settings.
3. Open Developer or Desktop app developer settings.
4. Click Edit Config if that button is available. This opens the Claude Desktop MCP config file.
5. Paste the JSON below, save the file, then fully quit and reopen Claude Desktop.

{
  "mcpServers": {
    "ifc-mcp": {
      "command": "npx",
      "args": ["-y", "ifc-mcp"]
    }
  }
}

If the file already contains JSON, add only the ifc-mcp entry inside the existing mcpServers object. Do not create a second mcpServers object.

Option 2: Open the config file directly

Use this only if you do not see Edit Config in Claude Desktop:

• Windows: %APPDATA%\Claude\claude_desktop_config.json
• macOS: ~/Library/Application Support/Claude/claude_desktop_config.json

Paste the same JSON from Option 1, save the file, then fully quit and reopen Claude Desktop.

Check

To check the setup, open Claude Desktop, start a new chat, click the + button near the chat box, and look for Connectors. IFC MCP should appear there after Claude Desktop has restarted. You should see the tools listed in the How It Works section above.

For more detail, see Claude's local MCP server guide.

Option 1: Use the install button

This opens VS Code and adds IFC MCP.

If VS Code asks whether you trust the server, accept it. Reload VS Code if the tools do not appear. Open Chat and use the tools/configure-tools button to see the IFC MCP tools.

Option 2: Add .vscode/mcp.json manually

{
  "servers": {
    "IFC MCP": {
      "type": "stdio",
      "command": "npx",
      "args": ["-y", "ifc-mcp"]
    }
  }
}

Reload VS Code after saving the file.

Open Chat and use the tools/configure-tools button to see the IFC MCP tools.

For more detail, see the VS Code MCP server docs.

Option 1: Use the Codex settings UI

1. Open VS Code.
2. Open the Codex panel.
3. Open Codex settings.
4. Select MCP servers.
5. Click Add server.
6. Enter:
   • Name: ifc-mcp
   • Command: npx
   • Arguments: -y ifc-mcp
7. Save the server.
8. Reload VS Code or start a new Codex session.
9. Check that the IFC MCP tools are available in the new Codex session.

Option 2: Use the terminal command

1. Open Terminal -> New Terminal in VS Code.
2. Paste and run:

codex mcp add ifc-mcp -- npx -y ifc-mcp

3. Reload VS Code or start a new Codex session.
4. Check that the IFC MCP tools are available in the new Codex session.

For more detail, see the Codex MCP docs.

Option 1: Use a project JSON file

1. Open your project in VS Code.
2. Create or edit .mcp.json in the project root.
3. Add this JSON and save the file:

{
  "mcpServers": {
    "ifc-mcp": {
      "command": "npx",
      "args": ["-y", "ifc-mcp"]
    }
  }
}

4. Reopen Claude Code in VS Code.
5. Type /mcp and approve the project server if Claude asks.
6. Check that ifc-mcp is connected.

Option 2: Use the terminal command for all projects

claude mcp add --transport stdio ifc-mcp --scope user -- npx -y ifc-mcp

Then reopen Claude Code in VS Code, type /mcp, and check that ifc-mcp is connected.

If the claude command is not found, install Claude Code first.

For more detail, see the Claude Code VS Code docs and Claude Code MCP docs.

Add this server entry to your MCP client config:

{
  "mcpServers": {
    "ifc-mcp": {
      "command": "npx",
      "args": ["-y", "ifc-mcp"]
    }
  }
}

Restart the MCP client after saving the config.

For more detail, see the Model Context Protocol docs.

Provide handlers that receive local file paths. Replace the functions below with your app's viewer actions:

import { createServer } from "ifc-mcp";
import { serveStdio } from "@modelcontextprotocol/server/stdio";

serveStdio(() => createServer({
  viewer: {
    "open-ifc-viewer": () => openThisViewer(),
    "show-ifc-file": ({ file_path }) => loadIfcIntoThisViewer(file_path),
    "set-bcf-view": ({ bcf_path }) => applyBcfToThisViewer(bcf_path),
    "clear-ifc-viewer": () => clearThisViewer(),
  },
}));

Return MCP tool results from your handlers: content contains feedback, and isError: true indicates failure.

Pass files selected in your app and callbacks for your existing viewer. Here, file is a selected IFC File, and viewer is your viewer instance:

import { createIfcMcpHost } from "ifc-mcp/browser";

const files = new Map([
  [file.name, { name: file.name, source: file }],
]);

const host = createIfcMcpHost({
  files,
  viewer: {
    "show-ifc-file": (name, bytes) => viewer.add(name, bytes),
    "set-bcf-view": (name, bytes) => viewer.add(name, bytes),
    "clear-ifc-viewer": () => viewer.clear(),
  },
});

const result = await host.handleToolCall({
  name: "show-ifc-file",
  input: { file_path: file.name },
});

File callbacks receive a filename and Uint8Array bytes. The BCF callback must return an MCP tool result describing whether the viewpoint was applied. Use host.tools for the tool definitions and host.getGeneratedFiles() for generated downloads. Call host.clearFiles() when those downloads are no longer needed.