401 Unauthorized / Unexpected Endpoint when connecting to LM Studio with Server Auth or MCP enabled

[almostfunctional]

Description: Recent updates to LM Studio (0.4.0+) enforce a strict token authentication scheme when Server Authentication or Model Context Protocol (MCP) tools are active. This causes Orpheus-FastAPI to crash with an Error: API request failed with status code 401 (invalid_api_key) because the outbound requests.Session doesn't pass a bearer token.

Additionally, if the base URL points strictly to /v1, LM Studio warns of an Unexpected endpoint or method (POST /v1) and fails to generate tokens.

Here is a tested walkthrough to modify the inference engine to dynamically inject the bearer token and fix the endpoint routing path.
Fix:

1. Generate your API token in LM Studio

2. Open: your Orpheus project directory and edit the inference engine file at: tts_engine/inference.py

3. Locate the while retry_count < max_retries: loop block.

Replace the standard session.post execution block with the following logic to handle the token handshake and endpoint appending:

    while retry_count < max_retries:
        try:
            # 1. Clone the headers dictionary and force inject the LM Studio Token with Bearer scheme
            auth_headers = HEADERS.copy() if isinstance(HEADERS, dict) else {}
            auth_headers["Authorization"] = "Bearer <YOUR_LM_STUDIO_TOKEN_HERE>"

            # 2. Fix the broken base path by forcing the legacy completions endpoint
            corrected_url = API_URL
            if corrected_url.endswith("/v1"):
                corrected_url = f"{corrected_url}/completions"
            elif corrected_url.endswith("/v1/"):
                corrected_url = f"{corrected_url}completions"

            # 3. Submit the network request to the proper authenticated endpoint
            response = session.post(
                corrected_url, 
                headers=auth_headers,
                json=payload,
                stream=True,
                timeout=REQUEST_TIMEOUT
            )
            
            if response.status_code != 200:
                # ... [The rest of your original streaming loop logic continues here unchanged] ...

Save inference.py and restart your Orpheus terminal application