In LM Studio, open the Developer tab and switch on Start server, or run lms server start in Terminal. LM Studio's docs say the server then listens on http://localhost:1234, and that an existing OpenAI client works with it once you point the client's base URL at http://localhost:1234/v1. By default LM Studio does not require authentication for API requests; you switch on Require Authentication in Server Settings and create a token. Its warning on serving beyond localhost is that any bind other than 127.0.0.1 exposes the server beyond localhost, so it recommends authentication first. Its system requirements page lists Apple silicon chips and macOS 14.0 or newer for Macs, and says Intel-based Macs are currently not supported. This page repeats those documents and ran nothing on a Mac.
localhost or on the network.lms, the command line tool that comes with the app: lms server start. The docs add that if LM Studio is not already running when you run an lms command, it starts running automatically.http://localhost:1234 once it is running, and that their examples assume port 1234; the port is a setting in Server Settings, and the developer page shows lms server start --port 1234.The lms page gives this sequence to download a model and serve it:
lms get openai/gpt-oss-20b
lms load openai/gpt-oss-20b
lms server start
LM Studio's OpenAI compatibility page lists six paths: /v1/models, /v1/responses, /v1/decisions, /v1/chat/completions, /v1/embeddings and /v1/completions. The docs say to switch an existing OpenAI client's base URL to http://localhost:1234/v1 and, for a request, to put the model identifier from LM Studio in the model field. They also list Anthropic-compatible endpoints and a REST API whose chat example posts to /api/v1/chat, and say Codex is supported because LM Studio implements POST /v1/responses. For Claude Code, see Claude Code with a local model.
Authorization: Bearer $LM_API_TOKEN.Server Settings has a Serve on Local Network switch. The docs say that once it is on, the server binds to a non-localhost address instead of localhost. Their warning, in full, reads: "Any bind other than 127.0.0.1 exposes the server beyond localhost; we recommend enabling authentication. Only do this if you know what you're doing." The Terminal form is:
lms server start --bind 0.0.0.0
The docs list three uses for it: other less powerful devices using a model on a more powerful machine, several people using one LM Studio, and IoT or edge devices calling the API.
mcp.json, which follows Cursor's notation: in the Program tab in the right-hand sidebar, click Install, then Edit mcp.json.mcp.json. The docs say it can be a security risk if those servers have access to your file system or private data, and that it requires Require Authentication to be enabled.mcp.json; the docs say those connections last only as long as the request and that only remote MCPs are supported.curl -fsSL https://lmstudio.ai/install.sh | bash, which runs a downloaded script; their page carries no warning for it. Start it with lms daemon up./v1/models returns every downloaded model and an inference call loads the model if it is not loaded; with it off, /v1/models returns only models loaded into memory, and you load one first. They also say JIT-loaded models are auto-unloaded by default after a period of inactivity.LM Studio's Offline Operation page says that in general LM Studio does not require the internet, and names chatting with models, chatting with documents and running a local server as examples. It lists five operations that do: searching for models, downloading models, the Discover tab's model catalog statistics, downloading runtimes and checking for app updates.
| Claim | Source |
|---|---|
Starting the server, the Developer tab, lms server start | LM Studio docs, Running the server |
| The settings: port, authentication, local network, MCP switches, CORS, JIT | LM Studio docs, Server settings |
| Default no authentication, tokens, the bearer header, 0.4.0 | LM Studio docs, Authentication |
Serve on Local Network, the bind warning, --bind 0.0.0.0 | LM Studio docs, Serve on local network |
| The OpenAI endpoints, the base URL, Codex | LM Studio docs, OpenAI compatibility |
MCP host, mcp.json, the warnings | LM Studio docs, Use MCP servers |
| llmster, run on login, JIT loading, auto-unload | LM Studio docs, Headless |
lms, port 1234, the get and load sequence, auto start | LM Studio docs, LM Studio, llmster and lms |
| Apple silicon, macOS 14.0, 16GB, Intel not supported | LM Studio docs, System requirements |
| What needs the internet | LM Studio docs, Offline operation |
lms server start in Terminal. LM Studio's docs say the server then listens on http://localhost:1234.Authorization: Bearer.lms server start --bind 0.0.0.0. The docs warn that any bind other than 127.0.0.1 exposes the server beyond localhost and recommend enabling authentication first.