n8n - 💡(How to fix) Fix Bug Report: mcpClientTool v1.2 opens persistent GET connection on HTTP Streamable transport [1 pull requests]

Describe the problem/error/question

The MCP Client node (@n8n/n8n-nodes-langchain.mcpClientTool v1.2) opens a persistent GET connection to the MCP server when using HTTP Streamable transport. This GET connection is never closed by n8n and has to be timed out by the server (600 seconds in our case). Once it times out, n8n immediately reconnects, creating a continuous loop of hanging GET connections that accumulates over time and causes the MCP server to become unresponsive.

What is the error message (if any)?

No error on n8n side. On the MCP server side:

GET /api/mcp 200 [ASYNC: canceled] 600083ms
MCP client connected: @n8n/n8n-nodes-langchain.mcpClientTool 1.2
GET /api/mcp 200 [ASYNC: canceled] 600069ms
MCP client connected: @n8n/n8n-nodes-langchain.mcpClientTool 1.2

This repeats indefinitely as long as the workflow is active.

Please share your workflow/screenshots/recording

Workflow has an AI Agent node with mcpClientTool v1.2 connected as a tool:

• Endpoint: https://metabase-instance-name/api/mcp
• Server Transport: HTTP Streamable
• Authentication: Header Auth

{ "nodes": [ { "parameters": { "options": {} }, "type": "@n8n/n8n-nodes-langchain.chatTrigger", "typeVersion": 1.1, "position": [3888, 1280], "id": "360337df-a246-4a82-95c0-b6f2e24b666f", "name": "Chat Trigger", "webhookId": "chat-trigger-webhook" }, { "parameters": { "options": { "systemMessage": "You are a data bot. Answer questions by querying the database via Metabase MCP tools. Never answer from memory." } }, "type": "@n8n/n8n-nodes-langchain.agent", "typeVersion": 3, "position": [4224, 1280], "id": "df446446-b420-47aa-bda7-573f83812c48", "name": "AI Agent1" }, { "parameters": { "endpointUrl": "https://metabase-instance/api/mcp", "authentication": "headerAuth", "include": "selected", "includeTools": ["search", "get_table_field_values", "get_table", "get_metric", "get_metric_field_values", "execute_query"], "options": {} }, "type": "@n8n/n8n-nodes-langchain.mcpClientTool", "typeVersion": 1.2, "position": [4400, 1504], "id": "93a10972-f316-466d-ac52-c0b67fff8e06", "name": "MCP Client", "credentials": {} }, { "parameters": {"options": {}}, "type": "@n8n/n8n-nodes-langchain.lmChatGoogleGemini", "typeVersion": 1, "position": [4192, 1584], "id": "73d3e82d-a0c7-437d-bacc-dd4bcdf4c446", "name": "Google Gemini Chat Model", "credentials": {} } ], "connections": { "Chat Trigger": {"main": [[{"node": "AI Agent1", "type": "main", "index": 0}]]}, "MCP Client": {"ai_tool": [[{"node": "AI Agent1", "type": "ai_tool", "index": 0}]]}, "Google Gemini Chat Model": {"ai_languageModel": [[{"node": "AI Agent1", "type": "ai_languageModel", "index": 0}]]} }, "pinData": {}, "meta": {"templateCredsSetupCompleted": true, "instanceId": "REDACTED"} }

Share the output returned by the last node

POST requests (actual tool calls) work correctly and respond in 1-3ms. The issue is the persistent GET connection running in the background, not the tool calls themselves.

Server logs show:

POST /api/mcp 200 2ms   ← tool calls, fine
GET  /api/mcp 200 [ASYNC: canceled] 600083ms  ← persistent keepalive, problem

Each GET holds a server thread for 600 seconds. With repeated n8n sessions, this caused our server's internal query queue to back up to 130+ requests making all queries time out.

Debug info

core

• n8nVersion: 1.123.43
• platform: docker (cloud)
• nodeJsVersion: 24.13.1
• nodeEnv: production
• database: sqlite
• executionMode: regular
• concurrency: 50
• license: enterprise (sandbox)

storage

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

pruning

• enabled: true
• maxAge: 720 hours
• maxCount: 25000 executions

client

• userAgent: mozilla/5.0 (macintosh; intel mac os x 10_15_7) applewebkit/537.36 (khtml, like gecko) chrome/148.0.0.0 safari/537.36
• isTouchDevice: false

Generated at: 2026-05-25T07:54:39.438Z

Expected behaviour: The MCP Client should not maintain a persistent GET connection when using HTTP Streamable transport.The MCP Client Tool opens persistent GET streams under HTTP Streamable transport and does not appear to close or reuse them correctly. Reconnection behavior causes accumulation of long-lived idle connections, eventually exhausting server-side resources.