Local LLM

You run a Local LLM MCP Server to connect local models with Claude Desktop and other MCP clients, keeping sensitive data on your machine while benefiting from cloud-compatible interactions. This setup gives you privacy-first processing, automatic model discovery, and a suite of tools to analyze, rewrite, and reason over content without leaving your device.

How to use

You start the server in either local (stdio) mode for Claude Desktop integration or in HTTP/HTTPS modes for broader network access. The server automatically detects models loaded in LM Studio, selects a default model, and exposes capability information for you to call from your MCP client. Use the local server for private reasoning, analysis, rewriting, and code review, then switch models or override per request as needed.

How to install

Prerequisites you need before installing:

• Node.js 18+
• LM Studio installed and at least one model loaded

Install and run steps you will follow in order:

git clone https://github.com/yourusername/local-llm-mcp-server.git
cd local-llm-mcp-server
npm install
npm run build

Configure Claude Desktop

Add a local MCP server entry to Claude Desktop by editing your Claude Desktop config to point to the local server. Use an absolute path to the built server file.

{
  "mcpServers": {
    "local_llm": {
      "command": "node",
      "args": ["/absolute/path/to/local-llm-mcp-server/dist/index.js"]
    }
  }
}

Start and use the server

Start Claude Desktop after configuring the local MCP server. The server will discover available models from LM Studio and expose a local capability surface you can query from Claude.

If you want to explore available models or capabilities, read the local resource paths exposed by the server, such as local://models and local://capabilities.

Convenience scripts

For easier server management, use the included scripts to start in different modes and manage certificates.

# Start in local mode (stdio - for Claude Desktop)
npm run start:local

# Start in remote mode (HTTP - for network access)
npm run start:remote

# Start in secure mode (HTTPS - encrypted network access)
npm run start:https

# Start in dual mode (stdio + HTTP for both local and remote)
npm run start:dual

# Generate SSL certificates for HTTPS
npm run generate:certs

# Stop all running servers
npm run stop

Remote Network Access

You can access the server from other devices on your home network or connect Claude Desktop remotely. Use HTTPS for secure remote access and, if needed during development, allow self-signed certificates for testing.

To connect Claude Desktop remotely, start HTTPS mode and add a remote MCP server entry with the proper URL and, if needed, an option to accept self-signed certificates.

# 1. Start HTTPS server
npm run start:https

# 2. Add to claude_desktop_config.json:
{
  "mcpServers": {
    "local_llm_remote": {
      "command": "npx",
      "args": ["-y", "mcp-remote", "https://localhost:3010/sse"],
      "env": {"NODE_TLS_REJECT_UNAUTHORIZED": "0"}
    }
  }
}

Security and privacy considerations

Your setup is privacy-first by design. All processing can stay on your machine, and you control privacy levels, telemetry, and data handling. Review and adjust privacy settings to match your requirements.

Configuration essentials

The server discovers models loaded in LM Studio automatically and exposes a local capability surface for use by MCP clients.

Important endpoints and capabilities are available through the local resources and the capability surface, which you can reference from your MCP client.

Troubleshooting

If you encounter connection issues with LM Studio, ensure LM Studio is running, the server has started, and the base URL matches your LM Studio configuration. If you experience performance problems, adjust the max concurrent requests or increase the request timeout. For privacy concerns, review privacy level settings and disable logging if handling confidential data.

Notes

This server is designed to work with local LLMs for privacy-sensitive tasks. Always review the privacy settings and ensure they meet your requirements before processing confidential data.

Available tools

local_reasoning

Use the local LLM for specialized reasoning tasks while keeping data private.

private_analysis

Analyze sensitive content locally without cloud exposure.

secure_rewrite

Rewrite or transform text locally for privacy.

code_analysis

Analyze code locally for security, quality, or documentation.

template_completion

Complete templates or forms using the local LLM.