Error messages

Unexpected endpoint or method.

Unexpected endpoint or method.

LM Studio's local server sends this back when a request hits a URL it doesn't serve. The caller gets a small JSON body, here from a request that left off /v1 (#618 on LM Studio's bug tracker):

HTTP/1.1 200 OK
X-Powered-By: Express
Content-Type: application/json; charset=utf-8

{"error":"Unexpected endpoint or method. (POST /chat/completions)"}

And LM Studio's server log shows the same thing with the method and path in brackets:

[ERROR] Unexpected endpoint or method. (GET /api/tags). Returning 200 anyway

That bracket is the useful part. It's the exact request LM Studio received, so you can compare it against the list of paths it serves. As of September 26, 2026, 15 issues on LM Studio's bug tracker quote the message.

Why your app didn't show an error

The status is 200, which means success in HTTP. A user filed that as a bug in December 2025 (#1323), asking for 404 or 405 instead, and later comments say it's the same on 0.4.13 and 0.4.18. Nobody from LM Studio had replied there by September 26, 2026. In practice your app thinks the call worked, tries to read a chat reply or a model list out of {"error": ...}, and fails with something unrelated. If an app says it connected but then finds no models or returns empty answers, check LM Studio's log for this line.

The paths LM Studio serves

LM Studio's docs list these, all on the one port (their examples use 1234):

• OpenAI-compatible: GET /v1/models, and POST to /v1/chat/completions, /v1/completions, /v1/embeddings and /v1/responses. The docs set the client's base URL to http://localhost:1234/v1.
• Anthropic-compatible: POST /v1/messages, with the base URL set to plain http://localhost:1234 (the Anthropic SDK adds /v1 itself).
• LM Studio's own REST API, released as v1 in 0.4.0: POST /api/v1/chat, GET /api/v1/models, and POST /api/v1/models/load, /unload and /download.
• The older REST API at /api/v0/..., which the docs still describe.

The mistakes behind most reports

Mixing the prefixes accounts for several. /chat/completions without /v1 fails (#618), and so does /api/v1/chat/completions, because the native API's chat path is /api/v1/chat. The reverse happens too: one user's POST /v1/models/unload failed until they added /api (#1680).

Apps written for Ollama are another source. They ask for Ollama's paths, like GET /api/tags, which LM Studio doesn't have. A request for Ollama-compatible endpoints has been open since November 2024 (#190). Pick the app's OpenAI-compatible provider option if it has one.

Some lines are harmless. Opening the server address in a browser logs GET /favicon.ico, and tools that poll their own health checks log paths like GET /api/health (reported from Opencode in #2085). If chat works, you can ignore those.

Reports that don't fit

Two open issues show paths the docs do list. #2085 logs GET /v1/models on LM Studio 0.4.17 in June 2026, and #2267 reports that the lmstudio/llmster-preview Docker image rejects /api/v1/chat and /api/v1/models while /v1/models works. Neither had an answer from LM Studio on September 26, 2026. On the Docker preview, the OpenAI-compatible /v1/ paths are the ones reported working.