LM Studio MCP Setup: Use MCP Servers With Local Models (mcp.json Guide)

LM Studio can act as an MCP host, so a model running on your own computer can call MCP tools. To add a server, open the Program tab in the right-hand sidebar and click Install > Edit mcp.json. Paste a server entry in the same mcpServers format Cursor uses, and save.

The file itself lives at ~/.lmstudio/mcp.json on macOS and Linux, and %USERPROFILE%/.lmstudio/mcp.json on Windows. LM Studio recommends editing it in the built-in editor. MCP support arrived in LM Studio 0.3.17 and covers both local and remote servers.

Checked against: LM Studio's MCP documentation, MCP deeplink documentation and 0.3.17 release post on 2 November 2026. eSEOspace did not test LM Studio.

Why use MCP with a local model?

Running a model in LM Studio keeps your prompts on your machine. Adding MCP lets that local model look things up or take actions. LM Studio's release post describes itself as an "MCP Host," the same role Claude Desktop plays. Keep in mind that a remote MCP server still sends data to its own service, even if the model is local. Our remote vs local MCP lesson explains the difference.

Step 1: Open mcp.json

1. Open LM Studio (0.3.17 or later).
2. Switch to the Program tab in the right-hand sidebar.
3. Click Install > Edit mcp.json.

The file opens in LM Studio's own editor. If you prefer your own editor, the 0.3.17 release post gives the paths:

Step 2: Add a server

LM Studio's docs use the Hugging Face MCP server as their example. It is a remote server, so it needs a URL and a token header:

{
  "mcpServers": {
    "hf-mcp-server": {
      "url": "https://huggingface.co/mcp",
      "headers": {
        "Authorization": "Bearer <YOUR_HF_TOKEN>"
      }
    }
  }
}

Replace <YOUR_HF_TOKEN> with your Hugging Face access token. Checked against LM Studio's docs on 2 November 2026.

A local server uses command and args in the same way as Cursor. For example, this standard entry runs the official Filesystem server:

{
  "mcpServers": {
    "filesystem": {
      "command": "npx",
      "args": ["-y", "@modelcontextprotocol/server-filesystem", "/Users/you/Documents"]
    }
  }
}

The release post says LM Studio starts "a separate process for each MCP server," and that servers relying on npx or uvx need those tools installed and on your system PATH. The MCP config generator can produce these blocks.

LM Studio's docs give one paste tip: "When adding MCP servers manually, copy only the content after "mcpServers": { and before the closing }." In other words, add the server entry inside the existing object rather than nesting a second mcpServers.

Step 3: Or use a one-click install link

Server publishers can offer an "Add to LM Studio" button. It uses a deeplink in this form:

lmstudio://add_mcp?name=hf-mcp-server&config=<base64-encoded JSON>

The config parameter is the server's JSON entry, Base64-encoded. LM Studio's deeplink page includes a generator. As with any one-click install, read what the server is before you click.

Step 4: Approve tool calls

The release post explains that when a model calls a tool, "LM Studio will show a confirmation dialog." You can review the arguments, edit them if needed, and choose to allow once or always. If you pick always, the dialog stops appearing for that tool. You can undo that later in App Settings, under the Tools & Integrations area.

Start with "allow once" for anything that writes files or sends data. Our permissions and approvals lesson explains why.

Watch your context window

LM Studio's docs warn that "some MCP servers were designed to be used with Claude, ChatGPT, Gemini and might use excessive amounts of tokens." This "may quickly bog down your local model and trigger frequent context overflows." Practical steps:

• Add one server at a time and test it.
• Prefer servers with a small, focused tool list.
• Load a model with a larger context length if your hardware allows it.

Use MCP through LM Studio's API

For developers, LM Studio's MCP via API docs say version 0.4.0 or newer can use MCP in API requests. Servers can be defined per request as "ephemeral" servers, or referenced from mcp.json.

Security notes

LM Studio's own docs put it bluntly: "Never install MCPs from untrusted sources." Some servers "can run arbitrary code, access your local files, and use your network connection." Check a server with our MCP server risk checklist before adding it.

Next step

If you need MCP for a team rather than one laptop, compare hosted clients in our best MCP clients roundup. For the basics of what a server can do, start with what is an MCP server.

Frequently asked questions