LMStudio-MCP

# Troubleshooting Guide This guide focuses on common issues encountered when using LMStudio-MCP, especially related to model compatibility. ## Connection Issues ### 404 Errors When Connecting to LM Studio If you encounter 404 errors when Claude attempts to connect to LM Studio: 1. Verify that LM Studio is running and has a model loaded 2. Check that the LM Studio server is running on port 1234 (default) 3. Try these fixes: - In `lmstudio_bridge.py`, change `http://localhost:1234/v1` to `http://127.0.0.1:1234/v1` - Restart LM Studio and ensure the server is running - Check your firewall settings to ensure the connection isn't blocked ## Model Compatibility Issues ### Problems with phi-3.5-mini-instruct_uncensored The phi-3.5-mini-instruct_uncensored model is known to have compatibility issues with the OpenAI chat completions API format. Here are some potential workarounds: 1. **Try a different model format**: If available, try using the non-uncensored version of phi-3.5-mini 2. **Adjust API parameters**: - Reduce the `max_tokens` value (try 256 or 512 instead of 1024) - Set `temperature` to a lower value (0.3-0.5) - Simplify your prompts to be more direct ### Other Model Compatibility Issues Different models might have different API requirements and constraints: 1. **Response format issues**: Some models don't properly format their responses according to the OpenAI chat API spec 2. **Token limitations**: Models might have different context window sizes 3. **System prompt handling**: Some models ignore or incorrectly process system prompts ## General Debugging If you need more detailed debugging information: 1. Run the bridge with detailed logging to stdout/stderr: ```bash python lmstudio_bridge.py 2>&1 | tee debug_log.txt ``` 2. Look for error messages in the LM Studio interface or logs 3. Try a simple test prompt first to verify basic functionality ## If All Else Fails If you encounter persistent issues with a specific model: 1. Try a different model that's known to work well with the OpenAI chat completions API 2. Check the LM Studio GitHub repository or forums for model-specific issues or recommendations 3. Consider contributing your findings to this repository to help others