n8n Just Released MCP - Is It The Future of AI & No Code?

Key Concepts

• MCP (Model Context Protocol): A way for large language models (LLMs) to communicate with everyday tools like Gmail, Google Calendar, Google Sheets, and Slack.
• MCP Client: A component within N8N that sends data to an external AI agent via the MCP server.
• MCP Server Trigger: A component within N8N that receives data from an external AI agent and triggers actions within N8N.
• External AI Agent: An LLM, such as Claude, that interacts with N8N through the MCP.
• SSE Endpoint: The URL used for communication between the MCP client and the MCP server trigger.
• Subflow: An alternative method within N8N to achieve similar functionality as the current MCP implementation.
• API Key: A security key required to access and use the OpenAI API.
• System Message: A message used to provide context to the AI model, such as the current date, to improve accuracy.

MCP Functionality in N8N: Client and Server Triggers

Naden has introduced MCP functionality with two key features: the MCP client and the MCP server trigger. These features facilitate communication between N8N and external AI agents like Claude.

How it Works:

1. A user sends a message to the AI agent through the N8N chat widget (e.g., "Add Roy with a budget of $500 to the Google Sheet").
2. The MCP client sends this data to the MCP server trigger via an SSE endpoint.
3. The MCP server trigger relays the data to the external AI agent (Claude).
4. Claude processes the request and sends instructions back to the MCP server trigger.
5. The MCP server trigger executes the instructions, such as updating a Google Sheet or scheduling a calendar event.

Example:

• The user asks Claude to schedule a coffee event for the next day at 2 PM.
• Claude requests permission, and upon approval, the event is added to the user's Google Calendar.

Key Features of MCP

• Simplified Tool Integration: MCP merges multiple tool functionalities into a single client, eliminating the need to create separate routes for each action (e.g., create, update, delete events in Google Calendar).
• Autogenerated Schema: MCP automatically generates the schema for tools, reducing boilerplate code and manual configuration. This includes URLs, headers, and JSON bodies for HTTP requests.
• Tool Discovery: MCP automatically discovers and integrates new API routes and functionalities, providing access to a wider range of tool capabilities without manual setup.

Limitations of Current Implementation

• Not True MCP: The current implementation primarily enables communication with external AI agents, rather than providing true MCP capabilities within N8N.
• Subflow Alternative: The functionality can be replicated using subflows within N8N, reducing the immediate benefits of the MCP implementation.
• Limited Use Cases: The speaker hasn't seen many business owners using this setup for their business.

Building the MCP Client and Server Trigger

The video demonstrates how to build the MCP client and server trigger from scratch within N8N.

MCP Client Setup:

1. Create a new workflow in N8N and name it "MCP Client."
2. Add an AI agent, which automatically includes a chat widget.
3. Add a chat model (e.g., OpenAI).
   • Requires an OpenAI API key with a minimum of $5 credit.
4. Add an MCP client tool.
   • Requires the SSE endpoint from the MCP server trigger.

MCP Server Trigger Setup:

1. Create a new workflow in N8N and name it "MCP Server Trigger."
2. Select "MCP Server Trigger" as the trigger.
3. Copy the generated URL (SSE endpoint).
4. Paste the URL into the SSE endpoint field in the MCP client.

Integrating with Claude:

1. Download and install the Claude desktop app.
2. Open Claude's settings and enable developer settings.
3. Edit the cloud_desktop_config.json file.
4. Replace the existing content with the provided JSON configuration (available in the video description).
5. Update the mcpURL field with the MCP server trigger URL.
6. Obtain an API key from N8N (Settings -> N8N API -> Create New API Key).
7. Update the accessToken field in the JSON configuration with the N8N API key, including "Bearer " before the key.
8. Save the cloud_desktop_config.json file.

Adding Tools to the MCP Server:

1. Add tools like Google Calendar to the MCP server trigger workflow.
2. Use AI to autofill the details for each tool.
3. Activate the MCP server trigger workflow.

Addressing Date Issues:

• AI models lack a concept of "now," leading to incorrect dates.
• Add a system message to the AI agent with the current date using the expression {{$now}}.

Takeaways and Future Potential

• The current MCP implementation in N8N is in its early stages and has limitations.
• It primarily facilitates communication with external AI agents, which can be achieved through subflows.
• The speaker believes that MCP has significant potential to streamline tool integration and automation in the future.
• As Naden continues to develop MCP, it is expected to become a valuable tool for businesses.
• The speaker provides access to a school community with courses, blueprints, and live calls for further assistance with AI and automation.

Notable Quotes

• "MCP is going to be gamechanging there's no doubt about it, it's just that the direct implementation as of today is not exactly where it could be."
• "...all of the things that I showed you could technically be just done through a subflow or it could pretty much be done as well through just the main AI agent..."

Conclusion

While the current MCP implementation in N8N has limitations and can be replicated using subflows, it represents a promising direction for the future of tool integration and automation. The ability to seamlessly connect LLMs with everyday tools has the potential to significantly streamline workflows and unlock new possibilities for businesses. As Naden continues to develop MCP, it is expected to become a valuable asset for users seeking to leverage the power of AI in their automation efforts.