Prerequisites
- Jan.ai 0.5+ — jan.ai
- A local model that supports tool use (function calling)
- Node.js 18+
- Agent MCP Studio at agentmcp.studio
Step-by-Step Setup
Get your relay URL
Open the studio → Settings → MCP Relay Bridge → copy URL → click Connect.
Download bridge.js
curl -O https://agentmcp.studio/bridge.js && npm install ws
-
Add MCP server in Jan.ai
In Jan.ai: Settings → Extensions → MCP → Add Server:
{ "name": "agent-mcp-studio", "command": "node", "args": ["/path/to/bridge.js", "wss://agentmcp.studio/api/relay/YOUR-UUID"] }Enable the server toggle. Jan.ai will list available tools from Agent MCP Studio.
Enable tools for your model
In Jan.ai's chat settings, toggle Tools on. Start a conversation — your local model can now call your browser-built tools.
Frequently Asked Questions
Yes — Jan.ai supports MCP as external tool providers. In Jan's Settings → Extensions → Tools, add your MCP server configuration. Jan's AI assistant can then call your Agent MCP Studio tools during chat.
Jan is a fully open-source, offline-first AI chat application. It runs LLMs locally on your hardware (CPU and GPU) with no cloud dependencies. Jan supports multiple models and has a growing extension ecosystem including MCP tools.
Jan itself runs offline, but the Agent MCP Studio relay bridge requires an internet connection to reach wss://agentmcp.studio/api/relay/.... However, once connected, your browser tools run entirely locally in Pyodide WASM.
Tool calling works with instruction-tuned models that support function calling — typically 7B+ models like Llama 3 Instruct, Mistral Instruct, Phi-3, and Qwen Instruct variants. Check the Jan model hub for tool calling support notes.
Both run models locally but differ in interface and ecosystem. Jan is fully open-source with a plugin system; LM Studio has a more polished GUI. Both support MCP tools via bridge.js the same way — the setup steps are nearly identical.