Hey! Loving the MCP server, but hit a snag with some clients (Hermes, and via a LiteLLM proxy); tool calls fail there while Claude Desktop works fine.
Looks like the tools declare an outputSchema but the responses never actually wrap the data into structuredContent, just plain text content. So strict clients reject it as not matching spec, while Claude just shrugs and reads the text anyway.
Anyways, thanks for the great tool!
Hi Boris,
Thanks very much for the feedback! Just fixed the outputSchema. Please let me know if there’s still an issue though.
This made me realize we’re lacking behind a few other more minor spec changes in MCP that we should bring up to date. Will address those too after the review of our newest version of our ChatGPT plugin has been approved by OpenAI. Hopefully soon 
Yep, crazy how fast specs are changing nowadays in this field!
Thanks for the quick fix, I just verified my clients are working all right!