MCP Servers para LM Studio
git-scarrow/mcp-servers-lmstudio
Bridge MCP com suporte a multi-agentes assíncronos para LM Studio
Ver no GitHubColeção de servidores MCP (Model Context Protocol) que habilitam modelos do LM Studio a colaborarem, consultarem uns aos outros e executarem operações assíncronas em paralelo. Inclui agentes multi-concorrentes, revisor de código, conectores para Oracle e MongoDB, com instaladores one-click.
Instalação
npx convez-ai git-scarrow/mcp-servers-lmstudio1 comando no terminal: detecta o formato do repositório (marketplace ou plugin avulso) e instala via Claude Code. Requer Node 18+ e o Claude Code instalado.
Métricas
Casos de uso
- Executar múltiplos modelos em paralelo com pool de conexões
- Revisar código automaticamente com modelos 30B+
- Conectar LM Studio a bancos de dados (Oracle, MongoDB)
- Orquestrar agentes especializados coordenados
- Gerenciamento automático de VRAM para carregamento de modelos
Veredito

Projeto robusto com instaladores one-click e suporte a multi-agentes concorrentes. A documentação é clara sobre requisitos de modelo (14B-32B), mas o repositório foi movido para Codeberg — verificar se ainda está mantido ativo.
README
A tradução para o português deste README ainda está na fila.
🚀 LM Studio MCP Servers
One-click installers for Model Context Protocol (MCP) servers that enable LM Studio models to collaborate, query each other, and work together through async operations.
🎯 Quick Install
Click any button below to install MCP servers directly into LM Studio:
Production Servers
| Server | Description | Install | Docs |
|---|---|---|---|
| Concurrent Multi-Agent | Query multiple models simultaneously with connection pooling | 🚀 Install | 📖 Docs |
| Code Reviewer | Automated code review using 30B+ models | 🚀 Install | 📖 Docs |
| Oracle Database | Connect LM Studio to Oracle databases | 🚀 Install | 📖 Docs |
| MongoDB | MongoDB operations from LM Studio | 🚀 Install | 📖 Docs |
Experimental Servers
| Server | Description | Install | Status |
|---|---|---|---|
| Agent Orchestrator | Coordinate multiple specialized agents | 🧪 Install | Beta |
| VRAM Manager | Auto-manage model loading within VRAM limits | 🧪 Install | Alpha |
💡 What is MCP?
Model Context Protocol (MCP) allows LM Studio models to:
- 🔄 Query other models - One model can ask another for help
- 🚀 Run concurrently - Multiple models working in parallel
- 🛠️ Use tools - Connect to databases, APIs, and services
- 🤖 Self-referential AI - Models can improve their own responses
📊 Model Requirements
Based on extensive testing, here are the minimum model requirements for effective tool use:
| Task Complexity | Minimum Model | VRAM | Use Cases |
|---|---|---|---|
| Simple | qwen2.5-coder-14b | 8.5GB | Basic function calls, single tools |
| Medium | codestral-22b | 13.5GB | Multi-step workflows, sequential logic |
| Complex | qwen3-coder-30b | 18GB | Nested workflows, dynamic selection |
| Expert | qwen3-32b | 19.5GB | Error handling, production systems |
🔧 Installation Methods
Method 1: One-Click Web Installer (Recommended)
- Click any install button above
- Your browser will open the installer page
- Click "Install in LM Studio"
- LM Studio will prompt to add the server
Method 2: Direct Deeplink
Copy and paste these links into your browser:
# Concurrent Multi-Agent Server
lmstudio://add_mcp?name=lmstudio-concurrent&config=eyJjb21tYW5kIjoicHl0aG9uMyIsImFyZ3MiOlsiL3BhdGgvdG8vbG1zdHVkaW9fYnJpZGdlLnB5Il0sImVudiI6eyJMTVNUVURJT19BUElfQkFTRSI6Imh0dHA6Ly9sb2NhbGhvc3Q6MTIzNC92MSJ9fQ==
Method 3: Manual Configuration
Add to ~/.lmstudio/mcp.json:
{
"mcpServers": {
"lmstudio-concurrent": {
"command": "python3",
"args": ["/path/to/lmstudio_bridge.py"],
"env": {
"LMSTUDIO_API_BASE": "http://localhost:1234/v1"
}
}
}
}
Method 4: CLI Installation
# Using the installer generator
python3 generate_installer.py --preset concurrent --open
# Or directly with LM Studio CLI
lmstudio mcp add-json 'lmstudio-concurrent' '{"command":"python3","args":["/path/to/server.py"]}'
🎨 Create Your Own MCP Server
Quick Start Template
# my_mcp_server.py
from mcp import server
import asyncio
@server.tool()
async def my_tool(param: str) -> str:
"""Your tool description"""
return f"Processed: {param}"
if __name__ == "__main__":
server.run()
Generate Installer
# Clone this repo
git clone https://github.com/samscarrow/mcp-servers-lmstudio.git
cd mcp-servers-lmstudio
# Generate installer for your server
python3 scripts/generate_installer.py \
--name "my-server" \
--script "/path/to/my_mcp_server.py" \
--output installers/my-server.html \
--open
📈 Performance Benchmarks
| Configuration | Response Time | Throughput | VRAM Usage |
|---|---|---|---|
| Single 14B Model | 2.3s | 26 tokens/s | 8.5GB |
| 2x 14B Concurrent | 2.4s | 52 tokens/s | 17GB |
| 4x 14B Concurrent | 2.8s | 95 tokens/s | 33GB* |
| 1x 30B Model | 4.1s | 18 tokens/s | 18GB |
*Using JIT loading with auto-unload
🛡️ Security
⚠️ Important: MCP servers can execute code on your system. Only install servers from trusted sources.
Each installer shows:
- Full configuration before installation
- File paths and permissions required
- Environment variables used
- Network endpoints accessed
🤝 Contributing
We welcome contributions! To add your MCP server:
- Fork this repository
- Add your server to
installers/ - Create documentation in
docs/ - Submit a pull request
Server Requirements
- Async/await support for concurrent operations
- Proper error handling
- Documentation with examples
- Security considerations documented
- Tested with LM Studio v0.3.0+
📚 Resources
📜 License
MIT License - See LICENSE file for details.
🙏 Acknowledgments
- LM Studio team for the amazing local LLM platform
- Anthropic for the Model Context Protocol specification
- Contributors who've shared their MCP servers
Made with ❤️ by the LM Studio community
Last updated: 2025-01-30