When using mcp please choose a suitable model like claude 3.7, it will save you more time!

The release of MCP Server Trigger by n8n is truly an excellent initiative. It unlocks greater possibilities for MCP and makes encapsulating MCP more convenient. I’ve been using it to build MCP over the past few days, but what drove me crazy was that it failed 9 out of 10 times when invoked in workflows. I spent days trying various solutions to fix this. I even raised the issue in the community, and the official team resolved it (they did a great job). However, when I tried to create new MCPs, I still encountered various errors—either from MCP itself or ChatModel. My patience was wearing thin. Until today, I decided to try switching models—guess what? All workflows ran perfectly, and all three MCPs executed flawlessly! The entire process completed in one go. The only change I made was switching to Claude 3.7. Here I must criticize all OpenAI models: whether it’s 4o, o4, or 4.1 (their naming conventions are worse than dogshit), they’re utterly terrible in MCP applications. While they might excel in other areas, Claude reigns supreme in MCP. So folks, if you encounter issues with n8n’s MCP, first throw OpenAI models into the trash and launch a missile at them.