Note: This is still under development and is a preview. Currently testing on Qwen3 Coder Next REAP 40B A3B from here https://huggingface.co/lovedheart/Qwen3-Coder-Next-REAP-40B-A3B-GGUF/resolve/main/Qwen3-Coder-Next-REAP-40B-A3B-Q4_K_XL.gguf

Qwen3 MCP Server

A complete MCP server that gives LM Studio's Qwen3 (or any local LLM) full coding agent capabilities including 80+ tools for file operations, command execution, git, web search, memory, planning, and a full skills system.

Features

• Two Server Modes:

  • HTTP Mode - Browser-based chat at http://localhost:3847/chat.html

  • MCP Mode - Direct integration with LM Studio's MCP interface

• 80+ Tools including:

  • File operations (read, write, edit, search, glob)

  • Command execution (shell, background processes)

  • Git operations (status, diff, commit, branch, push, pull)

  • Web tools (search, image search, URL fetch)

  • Memory & planning (notes, scratchpads, task lists, plans)

  • Media (read images, PDFs, screenshots)

  • Jupyter notebook support

  • ComfyUI integration (32 workflow tools)

  • GitHub Blog (Jekyll blog creation, themes, deployment)

• Tool Name Aliases - Server-side routing handles common model hallucinations (e.g., edit -> edit_file, bash -> execute_command)

• Skills System - 16 pre-installed skills + install more from awesome-agent-skills

Related MCP server: Friday MCP Server

Quick Start

HTTP Mode (Browser Chat)

# Start the server
start-chat.bat

# Open in browser
http://localhost:3847/chat.html

MCP Mode (LM Studio Direct)

Add to your LM Studio MCP config (~/.lmstudio/mcp-servers.json):

{
  "mcpServers": {
    "qwen3-mcp": {
      "command": "node",
      "args": ["<YOUR_PATH>/src/index.js"],
      "cwd": "<YOUR_PATH>"
    }
  }
}

Replace <YOUR_PATH> with the actual path where you installed the server.

Then restart LM Studio.

Installation

Prerequisites

• Node.js 18+

• LM Studio with a model loaded (Qwen3 recommended)

Quick Setup (Windows)

cd <YOUR_PATH>
setup.bat

This will:

1. Install npm dependencies

2. Create required directories

3. Configure LM Studio MCP automatically

Manual Setup

# Install dependencies
npm install

# Setup MCP config for LM Studio
npm run setup

npm Scripts

Project Structure

qwen3-mcp/
├── src/
│   ├── index.js           # MCP server (stdio mode)
│   └── tools/             # Tool implementations
│       ├── filesystem.js  # File operations
│       ├── edit.js        # File editing
│       ├── bash.js        # Command execution
│       ├── git.js         # Git operations
│       ├── search.js      # Glob/grep search
│       ├── web.js         # Web search, fetch
│       ├── memory.js      # Notes, scratchpads
│       ├── planning.js    # Plans, task tracking
│       ├── tasks.js       # Todo lists
│       ├── thinking.js    # Reasoning tools
│       ├── context.js     # Conversation context
│       ├── interaction.js # User prompts
│       ├── media.js       # Images, PDFs
│       ├── notebook.js    # Jupyter support
│       ├── comfyui.js     # ComfyUI workflows
│       ├── github-blog.js # Jekyll blog tools
│       └── skills.js      # Skills system
├── frontend/
│   ├── server.js          # HTTP server (port 3847)
│   ├── chat.html          # Browser chat interface
│   └── index.html         # Image viewer interface
├── skills/                # Installed skills (16)
│   ├── chrome-extension/  # Chrome extension development (MV3)
│   ├── code-review/       # Code review methodology
│   ├── comfyui-nodes/     # ComfyUI custom node development
│   ├── comfyui-workflow/  # ComfyUI workflow creation
│   ├── differential-review/ # Security-focused diff review
│   ├── docx/              # Word document creation
│   ├── frontend-design/   # Frontend UI/UX
│   ├── github-blog/       # Jekyll blog for GitHub Pages
│   ├── mcp-builder/       # Build MCP servers
│   ├── modern-python/     # Python tooling (uv, ruff)
│   ├── react-best-practices/ # React patterns (Vercel)
│   ├── shadcn-ui/         # Modern component library
│   ├── static-analysis/   # CodeQL, Semgrep, SARIF
│   ├── testing-handbook-skills/ # Fuzzers, sanitizers
│   ├── web-artifacts-builder/ # HTML/React prototypes
│   └── web-design-guidelines/ # UI/UX fundamentals
├── start-chat.bat         # Start HTTP server
├── stop-chat.bat          # Stop server
├── restart-chat.bat       # Restart server
└── install-skill.bat      # Install skills from GitHub

Available Tools (80+)

File Operations

Edit Tools

Search

Command Execution

Git

Web

Memory & Planning

GitHub Blog

Skills

Utilities

Skills System

Skills are instruction packages that teach the AI specialized tasks. The model auto-detects which skill to load based on your request.

Installed Skills (16)

Code Quality & Security:

• code-review - Code review methodology

• differential-review - Security-focused diff review

• static-analysis - CodeQL, Semgrep, SARIF

• testing-handbook-skills - Fuzzers, sanitizers, coverage

Web Development:

• react-best-practices - React patterns (Vercel)

• web-design-guidelines - UI/UX fundamentals

• shadcn-ui - Modern component library

• frontend-design - Frontend UI/UX

• web-artifacts-builder - HTML/React prototypes

ComfyUI & Creative:

• comfyui-nodes - Custom node development (V1 + V3 API)

• comfyui-workflow - Workflow creation (SD1.5/SDXL/SD3.5/Flux)

Development Tools:

• chrome-extension - Chrome extension development (MV3)

• mcp-builder - Build MCP servers

• modern-python - Python tooling (uv, ruff, pytest)

• docx - Word document creation

• github-blog - Jekyll blogs for GitHub Pages

Install More Skills

install-skill.bat https://github.com/anthropics/skills/tree/main/skills/pptx

Using Skills

Skills auto-load when they match your request. You can also load them manually:

"What skills do I have?"
"Load the comfyui-nodes skill"
"Use the docx skill to create a report"

Configuration

HTTP Server Port

Edit frontend/server.js:

const PORT = 3847;

Environment Variables

LM Studio Setup (HTTP Mode)

1. Load a model (Qwen3 recommended)

2. Enable API server (default: localhost:1234)

3. Open http://localhost:3847/chat.html

4. Enter API token if authentication is enabled

LM Studio Setup (MCP Mode)

1. Add MCP config (see Quick Start above)

2. Restart LM Studio

3. Tools appear automatically in model context

API Endpoints (HTTP Mode)

Testing

# Test a tool
curl -X POST http://localhost:3847/tool \
  -H "Content-Type: application/json" \
  -d '{"name":"get_current_time","args":{}}'

# List skills
curl http://localhost:3847/skills

Troubleshooting

Server Issues

# Check if running
curl http://localhost:3847/skills

# Restart
restart-chat.bat

# Force stop
stop-chat.bat

MCP Connection Issues

1. Check LM Studio console for errors

2. Verify path in mcp-servers.json is correct

3. Run node src/index.js manually to test

read_file Timeouts / WebSocket Errors

The read_file tool defaults to 500 lines to prevent LM Studio WebSocket timeouts. For large files, use pagination:

read_file with offset=1, limit=100    # Lines 1-100
read_file with offset=101, limit=100  # Lines 101-200

Tool Name Errors

If the model calls wrong tool names (e.g., edit instead of edit_file), the server has built-in aliases that route common mistakes. If you still see errors, update the system prompt in LM Studio — see SYSTEM_PROMPT.md for the correct prompt.

Tool Errors

1. Check browser console (F12) for HTTP mode

2. Verify working directory permissions

3. Check LM Studio model supports function calling

License

MIT