Choose the integration that does the job you need. A web search plugin can find public pages, a file tool can read a folder you give it access to, and a browser tool can interact with a page. Each needs different access. If you've already chosen an MCP server, follow the MCP setup guide for LM Studio.
Three ways to connect LM Studio
- An LM Studio plugin runs as an extension to the app. For example, the community DuckDuckGo plugin adds a search tool to chat. Other plugins can change the input that reaches the model, so check what the plugin does before enabling it.
- A Model Context Protocol (MCP) server is a separate program that provides tools. LM Studio is the host, and its MCP client connects to the server. The server can run on your computer or remotely. A local server can still make network requests.
- Another app using the local API sends requests to LM Studio to generate an answer. The API is a connection other programs can call. For example, a writing app that supports LM Studio's OpenAI-compatible API can use your local model from its own interface.
Pick a tool by the job you want to do
| Integration and job | Access required | First check |
|---|---|---|
| DuckDuckGo plugin to find public web pages | Sends search queries to DuckDuckGo. Keep private names and excerpts out of the query. | Ask for a link to public documentation you already know. Inspect the search terms, then open the result yourself. |
| Filesystem MCP to read or work with files | Access to the allowed folders. The server can also write and move files when those tools are enabled. Start with copies in a disposable folder. | Ask it to read one named text file and return a known phrase. Check the requested path and the answer. |
| Hugging Face MCP to find models and datasets | Sends lookups to Hugging Face's remote server. Check which account and tools are enabled before using account-specific features. | Look up a known public model or dataset by name and open its listing to confirm the match. |
| GitHub MCP to read repository information | Uses GitHub access granted by your sign-in or token. Start in read-only mode, limit toolsets to the job, and use credentials scoped to the repositories and permissions needed. | Ask for the title of a known issue in a public or test repository and compare it with GitHub. |
| Playwright MCP to navigate or interact with pages | Controls a browser and can act in any session available to that browser. Start with a separate test profile signed out of accounts. | Open a public documentation page and ask for a heading. Check the page that actually opened. |
A search result points you to a page. Open the source before relying on its content. Playwright can navigate pages, click controls, and fill forms. A separate browser profile keeps your everyday sessions out of the exercise, and the tool can still act on what that profile can reach. Playwright MCP does not provide a security sandbox.
Check access and try one integration at a time
For Filesystem, point the server at a folder with a copy of check.txt containing the words "purple marker". Enable its read tool, ask for the exact phrase and inspect the path it requested. Confirm the allowed directory before opening your real files.