Skip to content

AI Client Configuration

BlenderForge works with any AI assistant that supports the Model Context Protocol (MCP). This comprehensive guide covers detailed configuration for all popular AI clients.


Table of Contents

  1. Overview
  2. Claude Desktop
  3. Claude Code
  4. ChatGPT Desktop
  5. VS Code + GitHub Copilot
  6. Cursor IDE
  7. Windsurf (Codeium)
  8. Zed Editor
  9. Continue.dev
  10. Google Antigravity
  11. Custom MCP Clients
  12. Verifying Connection
  13. Troubleshooting

Overview

What is MCP?

The Model Context Protocol (MCP) is an open standard that defines how applications share context with large language models (LLMs). Think of it as a "USB-C port for AI applications" - it provides a standardized way to connect AI models to external tools and data sources.

Key Concepts: - MCP Server: Exposes tools and resources (BlenderForge is an MCP server) - MCP Client: AI applications that connect to servers (Claude, ChatGPT, etc.) - Transport: Communication method (stdio for local, HTTP for remote) - Tools: Functions the AI can invoke (like get_scene_info, execute_blender_code)

Universal Configuration Pattern

All MCP clients use a similar JSON configuration:

{
  "mcpServers": {
    "blenderforge": {
      "command": "blenderforge"
    }
  }
}

The differences are where this configuration goes and what format each client expects.

Prerequisites

Before configuring any AI client:

  1. Install BlenderForge: pip install blenderforge
  2. Verify installation: blenderforge --version
  3. Ensure command is in PATH: which blenderforge (macOS/Linux) or where blenderforge (Windows)

Claude Desktop

Developer: Anthropic Platform: macOS, Windows, Linux MCP Support: Full native support Documentation: Claude MCP Guide

Configuration File Location

Operating System Path
macOS ~/Library/Application Support/Claude/claude_desktop_config.json
Windows %APPDATA%\Claude\claude_desktop_config.json
Linux ~/.config/Claude/claude_desktop_config.json

Step-by-Step Setup

Step 1: Locate or Create the Configuration File

macOS:

# Create directory if it doesn't exist
mkdir -p ~/Library/Application\ Support/Claude

# Open or create the config file
open -e ~/Library/Application\ Support/Claude/claude_desktop_config.json

Windows (PowerShell):

# Create directory if it doesn't exist
New-Item -ItemType Directory -Force -Path "$env:APPDATA\Claude"

# Open or create the config file
notepad "$env:APPDATA\Claude\claude_desktop_config.json"

Linux:

# Create directory if it doesn't exist
mkdir -p ~/.config/Claude

# Open or create the config file
nano ~/.config/Claude/claude_desktop_config.json

Step 2: Add BlenderForge Configuration

Add the following JSON to your config file:

{
  "mcpServers": {
    "blenderforge": {
      "command": "blenderforge"
    }
  }
}

If you have other MCP servers, add BlenderForge to the existing mcpServers object:

{
  "mcpServers": {
    "filesystem": {
      "command": "npx",
      "args": ["-y", "@modelcontextprotocol/server-filesystem", "/Users/you/Documents"]
    },
    "blenderforge": {
      "command": "blenderforge"
    }
  }
}

Step 3: Restart Claude Desktop

  1. Quit Claude Desktop completely (not just close the window)
  2. macOS: Cmd+Q or Claude menu → Quit
  3. Windows: Right-click system tray icon → Exit
  4. Reopen Claude Desktop
  5. Look for the MCP indicator in the bottom-right corner of the chat input

Step 4: Verify Tools are Available

In Claude Desktop, you should see a hammer icon or "Tools" indicator. Click it to see available BlenderForge tools like: - get_scene_info - get_object_info - execute_blender_code - get_viewport_screenshot

Alternative: Desktop Extensions (Newer Method)

Claude Desktop now supports Desktop Extensions for easier MCP server management:

  1. Navigate to Settings → Extensions
  2. Click "Browse extensions"
  3. Search for MCP servers and click to install

Note: BlenderForge may need to be added manually until it's available in the extension marketplace.


Claude Code

Developer: Anthropic Platform: CLI tool (all platforms) MCP Support: Full native support Documentation: Claude Code MCP Configuration

Configuration

Claude Code uses the same configuration file as Claude Desktop. The configuration location depends on your operating system (see Claude Desktop section above).

Step-by-Step Setup

Step 1: Create/Edit Configuration

# macOS/Linux
mkdir -p ~/.config/Claude
cat > ~/.config/Claude/claude_desktop_config.json << 'EOF'
{
  "mcpServers": {
    "blenderforge": {
      "command": "blenderforge"
    }
  }
}
EOF

Step 2: Restart Claude Code

# Exit current session and restart
exit
claude

Step 3: Verify

# In Claude Code, ask:
"What MCP tools do you have available?"

ChatGPT Desktop

Developer: OpenAI Platform: macOS, Windows, Web MCP Support: Full support with Developer Mode Documentation: ChatGPT Developer Mode

Requirements

  • ChatGPT Plus, Pro, Business, Enterprise, or Education subscription
  • Developer Mode enabled

Step-by-Step Setup

Step 1: Enable Developer Mode

  1. Open ChatGPT Desktop or web app
  2. Go to Settings (gear icon)
  3. Navigate to Connectors → Advanced
  4. Toggle Developer Mode to ON

For Business/Enterprise users: - Go to Workspace Settings → Permissions & Roles - Enable Developer mode / Create custom MCP connectors

Step 2: Create a Custom Connector

  1. Go to Settings → Connectors → Create
  2. Fill in the connector details:
Field Value
Connector Name BlenderForge
Description Control Blender through natural language. Create 3D scenes, materials, and animations.
Command blenderforge
Transport stdio
  1. Click Save

Step 3: Use Developer Mode in Chat

  1. Open a new chat
  2. Click the Plus menu (or model selector)
  3. Select Developer mode
  4. Toggle BlenderForge connector ON

Note: The input field border will turn orange, indicating Developer Mode is active.

Step 4: Security Considerations

  • Custom MCP servers are third-party services not verified by OpenAI
  • Developer Mode allows read and write operations
  • ChatGPT will show confirmation modals for write actions
  • Report malicious servers to security@openai.com

VS Code + GitHub Copilot

Developer: Microsoft/GitHub Platform: VS Code extension MCP Support: Generally available (VS Code 1.99+) Documentation: VS Code MCP Servers

Requirements

  • VS Code 1.99 or later
  • GitHub Copilot extension installed
  • MCP enabled via chat.mcp.enabled setting (enabled by default)

Step-by-Step Setup

  1. Open VS Code
  2. Press Cmd+Shift+P (macOS) or Ctrl+Shift+P (Windows/Linux)
  3. Type "MCP: Add Server" and select it
  4. Choose "Add User Server" (global) or "Add Workspace Server" (project-specific)
  5. Enter the configuration:
{
  "blenderforge": {
    "command": "blenderforge"
  }
}

Method 2: Manual Configuration

Global Configuration (applies to all projects):

  1. Run command: "MCP: Open User Configuration"
  2. This opens ~/.vscode/mcp.json
  3. Add BlenderForge:
{
  "servers": {
    "blenderforge": {
      "command": "blenderforge"
    }
  }
}

Workspace Configuration (project-specific):

  1. Create .vscode/mcp.json in your project root
  2. Add the configuration:
{
  "servers": {
    "blenderforge": {
      "command": "blenderforge"
    }
  }
}

Step 3: Start the Server

  1. Open the .vscode/mcp.json file
  2. Click the "Start" button that appears at the top
  3. Or run command: "MCP: Start Server"

Step 4: Use in Agent Mode

  1. Open GitHub Copilot Chat panel
  2. Select Agent Mode (not Chat mode)
  3. Ask: "List my Blender scene objects"

Enterprise Considerations

  • Enterprises can control MCP usage via the "MCP servers in Copilot" policy
  • Policy is disabled by default for enterprise accounts
  • Admins can whitelist specific MCP servers

Cursor IDE

Developer: Cursor Platform: macOS, Windows, Linux MCP Support: Full native support Documentation: Cursor MCP Docs

Configuration File Locations

Scope Location
Global (all projects) ~/.cursor/mcp.json
Local (current project) .cursor/mcp.json

Step-by-Step Setup

Method 1: GUI Setup

  1. Open Cursor
  2. Press Cmd+Shift+P (macOS) or Ctrl+Shift+P (Windows/Linux)
  3. Type "MCP" and select "View: Open MCP Settings"
  4. Click Tools & Integrations in settings navigation
  5. Under MCP Tools, click New MCP Server
  6. Cursor creates an mcp.json file - add BlenderForge configuration

Method 2: Manual Configuration

Global Configuration:

# Create config directory
mkdir -p ~/.cursor

# Create/edit config file
cat > ~/.cursor/mcp.json << 'EOF'
{
  "mcpServers": {
    "blenderforge": {
      "command": "blenderforge"
    }
  }
}
EOF

Project-specific Configuration:

# In your project directory
mkdir -p .cursor

cat > .cursor/mcp.json << 'EOF'
{
  "mcpServers": {
    "blenderforge": {
      "command": "blenderforge"
    }
  }
}
EOF

Step 3: Restart Cursor

Close and reopen Cursor, or reload the window.

Step 4: Use MCP Tools

  1. Open the AI pane/chat panel
  2. Select Agent mode
  3. Cursor will prompt for tool approval each time (or enable Yolo mode for auto-approval)

Supported Transport Types

Cursor supports: - Local Stdio: Runs on your machine, streams via stdin/stdout - Streamable HTTP: Remote server handling multiple connections


Windsurf (Codeium)

Developer: Codeium Platform: macOS, Windows, Linux MCP Support: Full native support (Cascade integration) Documentation: Windsurf MCP Guide

Configuration File Location

Operating System Path
macOS/Linux ~/.codeium/windsurf/mcp_config.json
Windows %USERPROFILE%\.codeium\windsurf\mcp_config.json

Step-by-Step Setup

Method 1: MCP Marketplace

  1. Open Windsurf
  2. Click the MCPs icon in the top-right menu of the Cascade panel
  3. Browse the marketplace for available servers
  4. For servers not in marketplace, use manual configuration

Method 2: Settings UI

  1. Press Cmd+, (macOS) or Ctrl+, (Windows/Linux) to open Settings
  2. Scroll to "Plugins (MCP servers)" under Cascade
  3. Add BlenderForge configuration

Method 3: Manual Configuration

# Create directory
mkdir -p ~/.codeium/windsurf

# Create config file
cat > ~/.codeium/windsurf/mcp_config.json << 'EOF'
{
  "mcpServers": {
    "blenderforge": {
      "command": "blenderforge"
    }
  }
}
EOF

Step 3: Verify Setup

  1. Go to Cursor Settings → MCP
  2. BlenderForge should appear as a registered MCP server
  3. Check the toggle to enable it

Important Limits

  • Cascade has a limit of 100 total tools at any time
  • You can toggle individual tools on/off in MCP settings
  • Check logs at ~/.codeium/windsurf/logs for troubleshooting

Team/Enterprise Controls

  • By default, team members can configure their own MCP servers
  • Admins can whitelist specific servers (blocks all non-whitelisted servers)

Zed Editor

Developer: Zed Industries Platform: macOS, Linux MCP Support: Native support via context_servers Documentation: Zed MCP Docs

Configuration

Zed uses the context_servers key in its settings.

Step-by-Step Setup

Step 1: Open Zed Settings

Press Cmd+, to open Zed settings (JSON format).

Step 2: Add BlenderForge Configuration

{
  "context_servers": {
    "blenderforge": {
      "command": {
        "path": "blenderforge",
        "args": [],
        "env": {}
      }
    }
  }
}

Alternative: Extension-based Configuration

Some MCP servers are available as Zed extensions:

  1. Open Command Palette (Cmd+Shift+P)
  2. Search for "zed: extensions"
  3. Browse available MCP server extensions

Current Limitations

  • Only prompts are supported (shown under / command)
  • Only one prompt argument is supported per prompt
  • HTTP-based MCP servers have limited support

Debugging Tips

  1. Test your MCP server with: npx @modelcontextprotocol/inspector
  2. Check server status in the Agent Panel settings
  3. Write logs to stderr - Zed surfaces these for debugging

Continue.dev

Developer: Continue Platform: VS Code extension MCP Support: Full support in Agent mode Documentation: Continue MCP Guide

Configuration Methods

Continue.dev supports two configuration approaches:

  1. Standalone YAML files in .continue/mcpServers/ directory
  2. JSON config files copied from other tools (Claude Desktop, Cursor, etc.)

Step-by-Step Setup

  1. Create the mcpServers directory:
mkdir -p .continue/mcpServers
  1. Create a YAML file for BlenderForge:
# .continue/mcpServers/blenderforge.yaml
name: BlenderForge MCP Server
version: 0.0.1
schema: v1
mcpServers:
  - name: blenderforge
    command: blenderforge
    args: []

Method 2: JSON Configuration (Compatible with other tools)

  1. Create the mcpServers directory:
mkdir -p .continue/mcpServers
  1. Copy a JSON config file:
{
  "mcpServers": {
    "blenderforge": {
      "command": "blenderforge"
    }
  }
}

Save as .continue/mcpServers/blenderforge.json

Important Notes

  • MCP can only be used in Agent mode in Continue.dev
  • Standalone YAML files require metadata fields: name, version, schema
  • Continue.dev auto-detects config files in the .continue/mcpServers/ directory

Google Antigravity

Developer: Google Platform: Web-based AI coding platform MCP Support: Native support

Configuration

Add to your Antigravity MCP configuration:

{
  "mcpServers": {
    "blenderforge": {
      "command": "blenderforge"
    }
  }
}

Steps

  1. Open Antigravity settings
  2. Navigate to MCP configuration section
  3. Add BlenderForge as a server
  4. Save and refresh

Custom MCP Clients

For any other MCP-compatible client, the general pattern is:

Configuration Template

{
  "mcpServers": {
    "blenderforge": {
      "command": "blenderforge",
      "args": [],
      "env": {}
    }
  }
}

Python Client Example

from mcp import ClientSession
from mcp.client.stdio import StdioServerParameters, stdio_client

async def connect_to_blenderforge():
    """Connect to BlenderForge MCP server."""
    params = StdioServerParameters(
        command="blenderforge",
        args=[]
    )

    async with stdio_client(params) as (read, write):
        async with ClientSession(read, write) as session:
            # Initialize the connection
            await session.initialize()

            # List available tools
            tools = await session.list_tools()
            print("Available tools:")
            for tool in tools.tools:
                print(f"  - {tool.name}: {tool.description}")

            # Call a tool
            result = await session.call_tool("get_scene_info", {})
            print(f"Scene info: {result}")

# Run the client
import asyncio
asyncio.run(connect_to_blenderforge())

TypeScript Client Example

import { Client } from "@modelcontextprotocol/sdk/client/index.js";
import { StdioClientTransport } from "@modelcontextprotocol/sdk/client/stdio.js";

async function main() {
  const transport = new StdioClientTransport({
    command: "blenderforge",
    args: [],
  });

  const client = new Client({
    name: "blenderforge-client",
    version: "1.0.0",
  }, {
    capabilities: {}
  });

  await client.connect(transport);

  // List tools
  const tools = await client.listTools();
  console.log("Available tools:", tools);

  // Call a tool
  const result = await client.callTool({
    name: "get_scene_info",
    arguments: {}
  });
  console.log("Result:", result);
}

main().catch(console.error);

Verifying Connection

After configuring your AI client, verify the connection works:

Step 1: Ensure Blender is Running

  1. Open Blender
  2. Ensure the BlenderForge addon is enabled
  3. Press N to open the sidebar
  4. Go to the BlenderForge tab
  5. Click "Connect to MCP server"
  6. Status should show "Connected"

Step 2: Verify in AI Client

Ask your AI assistant any of these questions:

"What BlenderForge tools do you have available?"
"What objects are in my Blender scene?"
"Take a screenshot of my Blender viewport"

Expected Tools

The AI should list tools including: - get_scene_info - Query scene information - get_object_info - Get object details - execute_blender_code - Run Python in Blender - get_viewport_screenshot - Capture viewport - search_polyhaven_assets - Search free assets - generate_material_from_text - AI material creation - create_from_description - NLP object creation


Troubleshooting

"Tools not appearing"

  1. Restart your AI client completely (quit and reopen)
  2. Verify JSON syntax using a validator like jsonlint.com
  3. Check config file path matches your OS
  4. Ensure blenderforge is in PATH:
    which blenderforge  # macOS/Linux
    where blenderforge  # Windows
    
  5. Check AI client logs for MCP-related errors

"Connection refused"

  1. Ensure Blender is running with addon enabled
  2. Click "Connect to MCP server" in Blender's BlenderForge panel
  3. Check port (default: 9876) isn't used by another app
  4. Restart the connection in Blender if needed

"Command not found: blenderforge"

# Reinstall BlenderForge
pip install --upgrade blenderforge

# Or install with explicit path
pip install --user blenderforge

# Find installation location
pip show blenderforge

Client-Specific Issues

Client Common Issue Solution
Claude Desktop Config not loading Check JSON syntax, restart app
ChatGPT Developer Mode not visible Requires Plus/Pro subscription
VS Code Tools not in Copilot Ensure Agent Mode (not Chat)
Cursor Server not starting Check .cursor/mcp.json path
Windsurf 100 tool limit Disable unused tools in settings
Zed Only prompts work HTTP transport has limited support

Debug Mode

Enable verbose logging for troubleshooting:

export BLENDERFORGE_DEBUG=true
blenderforge

Check logs for detailed error messages.


Quick Reference Table

AI Client Config Location Format Key Name
Claude Desktop ~/Library/Application Support/Claude/ JSON mcpServers
ChatGPT Settings UI UI wizard Connector
VS Code Copilot .vscode/mcp.json JSON servers
Cursor ~/.cursor/mcp.json JSON mcpServers
Windsurf ~/.codeium/windsurf/mcp_config.json JSON mcpServers
Zed Settings JSON JSON context_servers
Continue.dev .continue/mcpServers/ YAML/JSON mcpServers

See Also


Sources