I’ve been testing the new beta functionality for the n8n instance-level MCP with OAuth enabled and was able to successfully connect from the ChatGPT UI.
From the documentation, it sounds like there should be ways to execute workflows as well. In the “ Accessing n8n MCP server | n8n Docs ” it says:
Search within workflows marked as available in MCP
Retrieve metadata and webhook information for workflows
Trigger and run workflows using webhook endpoints
However, as far as I can tell, only two tools are currently available:
search_workflows
get_workflow_details
I know the feature is still in beta, but is an “execute workflow” tool expected to be added soon, or am I missing something? From what I understand, the current implementation requires workflows to be exposed so that LLM clients can trigger them through another HTTP-request tool or something similar.
You’re correct that the current n8n instance-level MCP implementation only provides `search_workflows` and `get_workflow_details` tools. For workflow execution, you’ll need to use the webhook URLs that `get_workflow_details` returns and make HTTP requests through your AI client’s HTTP request capabilities.
To achieve your goal of executing private workflows through MCP:
• Enable MCP access for specific workflows in your n8n instance settings
• Use `get_workflow_details` to retrieve the webhook URL for each workflow you want to execute
• Have your AI client (ChatGPT) make HTTP POST requests to those webhook URLs when needed
This approach keeps workflows private while allowing controlled execution through the MCP interface. The [docs.n8n.io](Accessing n8n MCP server | n8n Docs) documentation suggests this is the intended workflow execution pattern for now, though a dedicated execution tool may be added in future releases.