Prerequisites
- LM Studio 0.3.5+ — lmstudio.ai
- A local model that supports tool calling (e.g. Qwen2.5, Mistral, Llama 3.1)
- Node.js 18+
- Agent MCP Studio at agentmcp.studio
Step-by-Step Setup
Get your relay URL and start bridge.js
Open the studio → Settings → MCP Relay Bridge. Copy your URL, click Connect, then:
curl -O https://agentmcp.studio/bridge.js && npm install ws node bridge.js wss://agentmcp.studio/api/relay/YOUR-UUID
-
Open LM Studio Developer settings
In LM Studio: Developer tab (left sidebar) → MCP Servers section → click Add MCP Server.
-
Configure the server
Fill in:
- Name: Agent MCP Studio
- Type: stdio
- Command: node
- Args:
/path/to/bridge.js wss://agentmcp.studio/api/relay/YOUR-UUID
Click Save. LM Studio will show the MCP server as connected.
Load a tool-capable model and chat
Load a model that supports function calling. In the chat, enable Tools mode. Your browser-built tools will appear in the tools list and the model can call them during conversation.
Qwen2.5-7B-Instruct, Mistral-7B-Instruct-v0.3, and Meta-Llama-3.1-8B-Instruct all have excellent tool-calling support. Smaller models (3B and below) may struggle with complex tool schemas.
Frequently Asked Questions
Yes — LM Studio added MCP support for local agents in 2025. Configure agent tool servers in LM Studio's settings and your locally-running LLMs can call Agent MCP Studio tools during conversations.
LM Studio is a popular desktop app for running large language models locally — Llama, Mistral, Phi, Gemma, and hundreds more. It provides a ChatGPT-like interface backed by your own hardware, with no cloud data sharing.
Yes — any model in LM Studio that supports tool/function calling (most 7B+ instruction-tuned models do) can use MCP tools. Quantized GGUF models work fine; check the model card to confirm tool calling support.
Yes — Agent MCP Studio works via the MCP relay bridge regardless of which LLM backend LM Studio is using. The bridge connects your browser tools to LM Studio's agent runtime, not directly to the model API.
Check: (1) LM Studio version supports MCP (update to latest), (2) the model you selected supports tool calling, (3) relay UUID matches Studio Settings, (4) bridge.js path is absolute and correct, (5) Node.js is installed.