LM Studio MCP Server Setup: Connect Tools to Local Models

LM Studio is a desktop application for running language models locally on your own hardware. It includes an AI chat interface and, in recent versions, support for MCP client connections that let a locally hosted model call outside tools during a session. Tools like fenbs can supply portable project context through an MCP connection so the local model starts each session with the rules, tasks, and current state of the work rather than from an empty prompt.

The practical gain from an MCP connection in LM Studio is the same as in any other AI client: the model can read from and write to outside services rather than relying only on what you type into the chat. The difference is that all inference runs on your local hardware, which matters for data that should not leave your machine.

Checking whether LM Studio supports MCP in your version

MCP client support in LM Studio was introduced in a specific release. Check the LM Studio changelog and release notes on the LM Studio website to confirm when MCP support was added and which features are included in your installed version. The feature may be in an early state in some releases and more complete in later ones.

Open the LM Studio settings or preferences to find the MCP or tools configuration section. If no such section exists in your version, update LM Studio before proceeding. Running an older version without MCP support means the configuration steps below will not apply.

Which models support tool use

Not every model that runs in LM Studio supports function calling or tool use. Tool use requires the model to produce structured outputs in a specific format that the application can parse and route to the correct MCP tool. Models trained or fine-tuned for tool use include recent versions of Mistral, Llama 3, Qwen, and others in the tool-use category on Hugging Face.

When selecting a model for use with MCP tools in LM Studio, look for models tagged as supporting function calling or tool use. Try a simple test before relying on the connection for real work: ask the model to call one of the configured tools with a known input and check that it returns a valid result rather than a plain text response.

Configuring an MCP server in LM Studio

LM Studio reads MCP server configuration from its settings interface or from a configuration file, depending on the version. Each server entry needs a name, the command to start the server process, and any required arguments or environment variables.

{
  "name": "fenbs",
  "command": "npx",
  "args": ["-y", "fenbs-mcp"]
}

Save the configuration, restart LM Studio if required, and open a chat session with a tool-capable model. Check whether the tools from the server appear in the session context. If the model does not call the tools, confirm it is loaded with a tool-use template and that the server started without errors.

Authentication compatibility for remote MCP servers

Local stdio MCP servers (started with a shell command) work with LM Studio in the same way they work with other clients. Remote HTTP MCP servers that require OAuth browser sign-in may need additional steps, because LM Studio's browser flow support depends on the version.

Check the LM Studio documentation for the current state of remote server and OAuth support. For servers that require a browser sign-in, you may need to complete the authentication step outside LM Studio and then provide the resulting token through an environment variable or configuration field that the server accepts.

Switching between LM Studio and cloud AI tools

One common pattern is using a local model in LM Studio for sensitive or offline work, and a cloud-hosted model for tasks that benefit from more capability or larger context windows. Each switch starts a new session in a different application. Without a shared record, the decisions and progress from the local session do not reach the cloud session automatically.

A project record outside both applications solves this. Both LM Studio and a cloud AI client read the same context over MCP, so the switch does not mean starting over. The current tasks, decisions, and Where we left off are the same in both sessions. See mcp-server-configuration for general configuration patterns that work across clients.

How to do this with fenbs

1. Create a free account and set up a project for the work you are doing with local models.
2. Follow the setup guide at Integrations to get the fenbs MCP connection details. Add the server entry to LM Studio's MCP configuration in its settings.
3. Load a tool-capable model in LM Studio and start a session. Ask it to read the project context before working on any task.
4. Record the decisions and rules that apply to this project so every session, whether in LM Studio or another AI tool, starts from the same approved baseline.
5. Before ending a session, ask the model to save Where we left off. Open the next session in any connected AI tool and ask it to read that note before starting.