I have set up an MCP Server in n8n with several tools connected to the trigger (as shown in the attached image). When a request is made, the execution starts and finishes as ‘Succeeded’ almost instantly (some runs in 12ms), but none of the connected tools are actually being invoked.
The server is reachable and the handshake seems to happen correctly, but the tool-calling logic is never triggered by the LLM or the MCP node. It’s a “silent success” where the workflow finishes without executing the tool nodes.
What is the error message (if any)?
There is no error message. The workflow status shows “Succeeded”, but no tools are called.
Please share your workflow
I can’t share the workflow by security issues. But I can show a print screen.
The output only shows the initial trigger data from the MCP request. There is no output data from tools like think, get_all, or find_documents because they are never executed.
Hi @Gregs, welcome to the n8n community!
When you say nothing changed, can you confirm the exact model name currently configured?
Also, could you check if the raw LLM response includes any tool_calls field? If the model response does not contain a tool call, that would explain the silent success.
Hi @Gregs Welcome to the community!
Although your setup looks nice, but this is not a good practice and passing all those tools would make every strong model hallucinate at some point, consider narrowing down your tool use case and better use an AI agent with proper context in system prompt, also just tighten your system prompt in that AI agent and everything should work fine.
Hi everyone, thanks for the replies! Let me clarify the timeline and the setup:
The Setup: I’m using MCP Servers connected to external MCP Clients, communicating with assistants powered mainly by Gemini and OpenAI models.
The Issue: Everything was working perfectly until yesterday at 3:00 PM. Suddenly, the assistants stopped responding correctly, and I noticed the tools were no longer being triggered.
The Update: I was 6 versions behind, so I decided to update to the N8N latest stable version to see if I could get better error logs or a fix.
Current Status: After the update, the MCP handshake is working again, but tool execution is now the bottleneck.
The Bug: I found that native nodes, like MongoDB, are not sending the required metadata/parameters correctly—specifically the supplyData method. Without this, the MCP client can’t ‘see’ or invoke the tool as it used to.
It seems the newer versions (or the updated MCP clients) are now strictly requiring this metadata, which some native apps nodes don’t provide out-of-the-box. I’ll try the sub-workflow (Call n8n Workflow Tool) as an adapter to see if that bridges the gap.
