Pull Request: Fix api_token not applied when using configure_default_client

Branch: fix/configure-default-client-api-token-global
Base: main
Target: lmstudio-ai/lmstudio-python

Summary

When calling configure_default_client(api_host, api_token=...) and then using the convenience API (e.g. list_downloaded_models()), the LM Studio server returns "An LM Studio API token is required to make requests to this server, but none was provided." The token is never sent because the module-level _default_api_token is never updated: only _default_api_host is declared global in configure_default_client, so the assignment _default_api_token = api_token creates a local variable and the global remains None. This pull request fixes that by declaring both globals.

Problem

Expected behavior (per Authentication docs)

• Call lms.configure_default_client(api_host, api_token="sk-lm-...") (or set LM_API_TOKEN).
• Subsequent calls such as lms.list_downloaded_models() use the default client and send the token so token-protected LM Studio servers accept the connection.

Actual behavior

• After configure_default_client(api_host, api_token="sk-lm-..."), the first use of the convenience API (e.g. list_downloaded_models() → get_default_client()) creates Client(_default_api_host, _default_api_token).
• The module-level _default_api_token was never set, because in configure_default_client only _default_api_host is declared global. The line _default_api_token = api_token therefore assigns to a local variable; the global _default_api_token stays None.
• The client is therefore created with api_token=None, and the websocket connects without token. The server responds with authentication failure.

Root cause

In src/lmstudio/sync_api.py, configure_default_client:

def configure_default_client(api_host: str, api_token: str | None = None) -> None:
    """Set the server API host for the default global client (without creating the client)."""
    global _default_api_host   # <-- only _default_api_host is global
    if _default_client is not None:
        ...
    _default_api_host = api_host
    _default_api_token = api_token   # <-- this assigns to a LOCAL variable

So _default_api_token is never written to the module global. Later, get_default_client() does:

_default_client = Client(_default_api_host, _default_api_token)  # _default_api_token is still None

Solution

Declare _default_api_token as global in configure_default_client so the assignment updates the module-level variable:

global _default_api_host, _default_api_token

One-line change in sync_api.py (around line 1712).

Testing

• Before fix: With a token-protected LM Studio server, configure_default_client(host, api_token=token) followed by list_downloaded_models() raises LMStudioServerError: Authentication failed: ... none was provided.
• After fix: The same sequence succeeds; the SDK sends the token and the server accepts the request.

Repro (with LM Studio running with API token enabled):

import lmstudio as lms
lms.configure_default_client("localhost:1234", api_token="sk-lm-xxxxxxxx:yyyyyyyyyyyyyyyyyyyy")
models = lms.list_downloaded_models("llm")  # Before fix: auth error. After fix: returns list.

Checklist (per CONTRIBUTING.md)

• Single-line change; no behavior change except fixing the bug.
• Aligns with documented Authentication API (api_token and LM_API_TOKEN).
• Branch is based on the main source
• Follows existing code style (no formatting changes).
• All tests pass (uv run pytest tests/ -m "not lmstudio" — 177 passed; full suite requires LM Studio with API token).
• Static checks pass (uv run ruff check src/ tests/ examples/plugins and uv run mypy --strict src/ tests/).