From Zero To SaaS #46 MCP Deep Dive

Developers like to standardize code to make it easier to connect different codebases, such as the most famous one for the REST API. However, for LLMs, Anthropic came up with MCP, the Model Context Protocol.

When you call the API of an LLM, you must define the user, content, model, and max tokens

MCP adds on top of that:

• tools. (Functions)
• resources (Context)
• prompt templates (Thinking Structure)

MCP has a Client that translates a standard language prompt into an MCP protocol request, which is sent to the Server, which processes that request, and then it sends it back to the client.

That Server is often local, which is good for data privacy, but can also be hosted online.

One downside of MCP is that it cannot be installed with a single click on the web version of Claude or ChatGPT; however, it is accessible to all serious LLM API providers and can be run on Claude Desktop.

If you want to use one button in the OpenAI environment, there is a workaround for MCP servers by connecting Zapier to the OpenAI Playground.

If you are a fan of Claude and want to get hands-on experience, take this introductory course.

The No Code Way (under 5 min)

You need to download Claude Desktop and an n8n account.

1. Create a workflow and add the MCP node

2. Add a functional tool; in this example, I pick Google Calendar to illustrate how to set up the settings.

A) Pick one calendar from the list

B) Add the example of how the time has to be written in the description.

You have to add in Additional Options, Summary, and add the description. This is the name of the event.

Hit the save button

3. Got to Claude Desktop, click on settings, go to developer, and click on Edit Config.

4. Add the production URL here, between the " ".

5. Save to the file, and quit Claude Desktop and open it again.

(In case it didn't work, watch this video)

6. Make sure to always allow.

Each time you have to fix a bug, save the version in the n8n workflow and restart Claude Desktop.

7. BONUS if you want to call another workflow, which can be an agent.

8. Bonus n8n make for you MCP server, which can be called by an AI agent within n8n.

Use cases:

1. ChatGPT subscription is about 27 euros, n8n is 24 euro, you still need to pay for OpenAI, but tokens are cheaper. N8N gives you also a UI + link not as great as ChatGPT though.

2. You can connect to all the tools on n8n and if a platform has API, you can connect it with HTTP tool.

• Contract/ Proposal Maker you connect the MCP with transcription tool.
• Data Analysis to Data Visualisation you can combine this with Artifacts of Anthropic.

Key Takeaways

1. If you use Claude this is amazing
2. If you use ChatGPT you can better learn how to build GPTS
3. If you don't like code, stick to n8n
4. Is this essential knowledge to know or learn outside GTM engineering, or ai agent building > No.