MCP tool calls are failing due to new toolCallId, previous n8n worfklows won't run on update to most recent version!

[bsk-amboss]

Describe the problem/error/question

MCP tool calls are failing due to new toolCallId, previous n8n worfklows won't run on update to most recent version!

What is the error message (if any)?

In one workflow I have this error:
[
{
"response":
[
{
"type":
"text",
"text":
"Error calling tool 'my_tool': 1 validation error for call[my_tool]\ntoolCallId\n Unexpected keyword argument [type=unexpected_keyword_argument, input_value='toolu_015kA1pdA84pTN4dfAkQ1XEg', input_type=str]\n For further information visit https://errors.pydantic.dev/2.11/v/unexpected_keyword_argument"
}
]
}
]

In another workflow, my agent just shows its running but no LLM or tools are called.

Please share your workflow/screenshots/recording

(Select the nodes on your canvas and use the keyboard shortcuts CMD+C/CTRL+C and CMD+V/CTRL+V to copy and paste the workflow.)
⚠️ WARNING ⚠️ If you have sensitive data in your workflow (like API keys), please remove it before sharing.

Share the output returned by the last node

Debug info

core

• n8nVersion: 1.119.1
• platform: docker (cloud)
• nodeJsVersion: 22.19.0
• database: sqlite
• executionMode: regular
• concurrency: 5
• license: community

storage

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

pruning

• enabled: true
• maxAge: 168 hours
• maxCount: 2500 executions

client

• userAgent: mozilla/5.0 (macintosh; intel mac os x 10_15_7) applewebkit/537.36 (khtml, like gecko) chrome/142.0.0.0 safari/537.36
• isTouchDevice: false

Generated at: 2025-11-11T08:28:14.370Z}

[n8n-assistant]

Hey @bsk-amboss,

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-5439".

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-5439

[AI-Guru]

I experience the same. The MCP Client tool sends the fields sessionId, action, chatInput, and toolCallId. This essentially breaks MCP support.

I could do a workaround for the MCP I am working on by adding the fields to my MCP tools, thus preventing Pydantic validation errors.

[garyellow]

I can confirm I'm encountering this issue as well. It appears the AI Agent is sending extraneous parameters (likely those passed to the parent node), which results in a parsing failure.

core

• n8nVersion: 1.119.1
• platform: k8s (self-hosted)
• nodeJsVersion: 22.19.0
• database: PostgreSQL
• executionMode: worker
• concurrency: 10
• license: community
• error node: MCP Client Tool Node
• error message:

Internal error: Error calling tool 'tavily_search': 2 validation errors for call[tavily_search]\n公司名稱\n  Unexpected keyword argument [type=unexpected_keyword_argument, input_value='台積電', input_type=str]\n    For further information visit https://errors.pydantic.dev/2.12/v/unexpected_keyword_argument\ntoolCallId\n  Unexpected keyword argument [type=unexpected_keyword_argument, input_value='8f0601b9-88de-457c-993f-79bf9169e2c0', input_type=str]\n    For further information visit https://errors.pydantic.dev/2.12/v/unexpected_keyword_argument

[garyellow]

Root Cause Analysis

I've investigated this issue and found the exact location causing the problem.
Cooperating with Claude 4.5 Sonnet.

🔍 The Problem

The issue originates from how n8n merges parent node (AI Agent) parameters with tool call arguments before passing them to MCP tools.

Location: packages/core/src/execution-engine/requests-response.ts (lines 65-76)

// Get the item index from action metadata to access the correct agent input data
const itemIndex = (action.metadata as { itemIndex?: number })?.itemIndex ?? 0;
// Use index 0 for main input as agents have only one main input connection
const agentInputData = executionData.data.main?.[0]?.[itemIndex];

// Merge agent's input data with action.input so tools have access to workflow data
// action.input takes precedence to allow tool-specific parameters to override
const mergedJson = {
    ...(agentInputData?.json ?? {}),  // ← ALL parent node parameters included here
    ...action.input,                   // ← Tool-specific parameters
    toolCallId: action.id,            // ← Internal tracking parameter
};

This code merges:

1. All input data from the AI Agent (agentInputData?.json) - including Agent-specific parameters like sessionId, action, chatInput, and user input fields
2. Tool-specific parameters (action.input)
3. Internal tracking parameter (toolCallId)

📍 Where It Breaks

Location: packages/@n8n/nodes-langchain/nodes/mcp/McpClientTool/McpClientTool.node.ts (lines 454-467)

const toolName = item.json.tool;
if (toolName === tool.name) {
    // Extract the tool name from arguments before passing to MCP
    const { tool: _, ...toolArguments } = item.json;  // ← Only filters out 'tool'
    const params: {
        name: string;
        arguments: IDataObject;
    } = {
        name: tool.name,
        arguments: toolArguments,  // ← Contains ALL merged parameters!
    };
    const result = await client.callTool(params, CallToolResultSchema, {
        timeout: config.timeout,
    });

The current code only filters out the tool property, but all other parameters from the parent Agent node are still passed to the MCP Server, causing Pydantic validation errors.

💥 Error Example

From the issue reports, we can see parameters that shouldn't be sent to MCP tools:

公司名稱 (user input to Agent)
  Unexpected keyword argument [input_value='台積電']
toolCallId (internal n8n parameter)
  Unexpected keyword argument [input_value='8f0601b9-88de-457c-993f-79bf9169e2c0']
sessionId (Agent parameter)
  Unexpected keyword argument
action (Agent parameter)
  Unexpected keyword argument
chatInput (Agent parameter)
  Unexpected keyword argument

✅ Proposed Solution

The MCP Client Tool should filter out all internal n8n parameters and parent node context before passing arguments to MCP tools:

// Define parameters that should not be passed to MCP tools
const INTERNAL_PARAMS = ['tool', 'toolCallId', 'sessionId', 'action', 'chatInput'];

const toolName = item.json.tool;
if (toolName === tool.name) {
    // Filter out internal n8n and parent node parameters before passing to MCP
    const toolArguments = Object.keys(item.json).reduce((acc, key) => {
        if (!INTERNAL_PARAMS.includes(key)) {
            acc[key] = item.json[key];
        }
        return acc;
    }, {} as IDataObject);
    
    const params = {
        name: tool.name,
        arguments: toolArguments,
    };
    // ... rest of the code
}

Alternatively, a more robust approach would be to only pass parameters that are explicitly defined in the tool's input schema, rather than filtering by a blocklist.

📋 Impact

This affects:

• All workflows using MCP Client Tool Node with AI Agent
• Any MCP server that uses strict Pydantic validation
• Workflows that were working in previous versions (breaking change in v1.119.1)

🔗 References

• Root cause: requests-response.ts:65-76
• Where it breaks: McpClientTool.node.ts:454-467
• Related test: requests-response.test.ts:27-68

[agaldemas]

facing the same issue, since last update...
Like @garyellow, I suggest to use the more robust approach to only pass parameters that are explicitly defined in the tool's input schema, rather than filtering by a blocklist.

[liaz-af]

I think this should be considered a security issue by n8n.io team.
The MCP server has no reason to know the user prompt, or any other parameter for that matter, other than the input parameters it defined and filled by the LLM.

This is why I believe the alternative suggested by @garyellow is the correct one - when I connected an "Edit Fields" node to the AI agent it passes to MCP client the parameters as received fro the "Edit Fields". This is of course a security issue

[RoyChang-NetBaseQuid]

We found an interesting behavior that the old workflows created before our n8n upgrading are using different version for AI Agent Node, and they are still working fine
So, a workaround is copy the node from those workflows
Or you can try copy and paste the following json into your workflow directly, I'm not sure does it work for everyone

{
  "nodes": [
    {
      "parameters": {
        "promptType": "define",
        "text": "={{ $json.prompt }}",
        "options": {
          "maxIterations": 3,
          "enableStreaming": true
        }
      },
      "type": "@n8n/n8n-nodes-langchain.agent",
      "typeVersion": 2.2,
      "position": [
        432,
        0
      ],
      "id": "7d4a7c24-72e8-4f66-b0b3-7ffa7fb572cd",
      "name": "AI Agent1"
    }
  ],
  "connections": {
    "AI Agent1": {
      "main": [
        []
      ]
    }
  },
  "pinData": {}
}

[liaz-af]

Any word if n8n.io sees this new behaviour as an issue?
We need to decide ASAP wether to wait or to change our tools

[agaldemas]

"typeVersion": 2.2

Thanks, it works fine with this AI agent node version !
no extra parameters !
But this behaviour should be reconducted in "typeVersion": 3 !

[skware]

Any update on this ? this makes n8n useless if you can't run mcp tools.

[krisn0x]

Thank you everyone, accepting routing this internally, so the team can take a look.