MCP Client not called in case of (additional) Text-Output of LLM

[ulftheo]

Describe the problem/error/question

In case the LLM is creating multiple outputs, the MCP Client / Tools is not invoked anymore. In more detail:
Output only consists of MCP Client Call? -> OK
Output consists of Text + MCP Client Call -> Not OK

What is the error message (if any)?

There's no error message, just the LLM enters a loop without MCP Function Call

Please share your workflow/screenshots/recording

Share the output returned by the last node

Not necessary

Debug info

Working Function call:

      "content": {
        "parts": [
          {
            "functionCall": {
              "name": "getAccessibleAtlassianResources",
              "args": {}
            },
            "thoughtSignature": "AAA"
          }
        ],
        "role": "model"
      },
      "finishReason": "STOP",
      "finishMessage": "Model generated function call(s)."
    }

MCP Function does not get invoked:

      "content": {
        "parts": [
          {
            "text": "Calling MCP_Client with input: {\"tool\":\"getJiraIssue\",\"cloudId\":\"XXX\",\"issueIdOrKey\":\"YYY-1234\"}"
          },
          {
            "functionCall": {
              "name": "MCP_Client",
              "args": {
                "tool": "getJiraIssue",
                "cloudId": "XXX",
                "issueIdOrKey": "YYY-1234"
              }
            },
            "thoughtSignature": "ZZZ"
          }
        ],
        "role": "model"
      },
      "finishReason": "STOP",
      "finishMessage": "Model generated function call(s)."
    }

core

• n8nVersion: 2.9.1
• platform: docker (self-hosted)
• nodeJsVersion: 24.13.1
• nodeEnv: production
• database: sqlite
• executionMode: regular
• concurrency: -1
• license: enterprise (production)

storage

• success: all
• error: all
• progress: false
• manual: true
• binaryMode: filesystem

pruning

• enabled: true
• maxAge: 336 hours
• maxCount: 10000 executions

client

• userAgent: mozilla/5.0 (macintosh; intel mac os x 10_15_7) applewebkit/537.36 (khtml, like gecko) chrome/145.0.0.0 safari/537.36
• isTouchDevice: false

Generated at: 2026-02-22T16:24:18.780Z}

[n8n-assistant]

Hey @ulftheo,

Thank you for reaching out! We’ve received your issue and are looking into it. To help us track this internally, we’ve created a Linear ticket with the reference: "GHC-6969".

We’ll keep you updated as we make progress, but if you have any additional details or context to share, feel free to add them here - it’s always helpful!

Thanks for your patience and for bringing this to our attention.

---

For INTERNAL use only: GHC-6969

[ulftheo]

Just to add, could be also related to the fact, that the Function call is now referring to "MCP_Client" instead of the function/Tools exposed by the MCP Client (e.g. getJiraIssue)? The input to the LLM on the 2nd iteration is as follows:

AI: Calling MCP_Client with input: {"tool":"getAccessibleAtlassianResources","id":"XXX"}\nTool:[{"response":[{"type":"text","text":"[{\"id\":\"XXXe\",\"url\":\".....

Maybe this is causing the behaviour?

[tamy-commits]

Hi @ulftheo, thank you for that extra bit of context!

Your observation about the functionCall referring to MCP_Client instead of the specific tool name is a very strong lead. To help us investigate this mapping logic:

• Could you please confirm if this still occurs on version 2.10.2? We've had several updates to AI Agent parsing recently
• Please share the JSON export of your workflow. We need to see exactly how the MCP Client and the AI Agent are connected to reproduce this multi-part response scenario.

#triage:needs-info