Make.com Just Released MCP–Learn Everything in 10 Mins

Key Concepts

• MCP (Model Context Protocol): An API standard for AI agents to communicate effectively and connect with various tools and applications.
• Tool Discovery: MCP's ability to automatically incorporate new tools and functionalities as soon as they are released by providers like Google Calendar.
• Merging Methods: MCP's feature to combine multiple actions (create, update, delete, search) for a single application into one module.
• Natural Language Processing: MCP's capability to understand and execute actions based on natural language input.
• MCP Server Trigger: A Make.com trigger that receives data from AI agents like Claude or ChatGPT.
• MCP Client Module: A Make.com module that allows interaction from Make.com to external AI agents (currently limited in functionality).
• OCR (Optical Character Recognition): The ability of a software to recognize text within an image or document.
• On-Demand Trigger: A type of trigger in Make.com that allows scenarios to be initiated by external requests, making them accessible as tools in AI agents.

Setting up MCP with Make.com, ChatGPT, and Claude

MCP Overview and Benefits

• MCP allows connecting AI models like ChatGPT, Claude, and Cursor to over 2,000 applications through Make.com.
• It standardizes communication between AI agents and tools, similar to how HTML standardizes web presentation or JSON standardizes data transfer.
• Key Features:
  • Tool Discovery: Automatically incorporates new tool functionalities without waiting for manual implementation.
  • Merging Methods: Combines multiple actions (e.g., create, update, delete) for a single application into one module, simplifying workflows.
  • Natural Language Processing: Understands and executes actions based on natural language input, eliminating the need for complex HTTP configurations.

Practical Example: Invoice Data Extraction

• The video demonstrates extracting data (name, total, tax) from an invoice using Claude and sending it to a Google Sheet via Make.com.
• Claude removes line items from the invoice and sends the relevant data to the MCP server, which then populates a Google Sheet.
• This showcases MCP's ability to process unstructured data and integrate it into structured applications.

Setting up MCP Access in Make.com

• Creating an MCP Token:
  1. Navigate to your profile in Make.com.
  2. Select "MCP Access."
  3. Create a new MCP token and name it (e.g., "MCP Server").
  4. Copy the generated unique key/URL. This URL is used to connect to OpenAI and Claude instances.
• Important: The trigger in Make.com needs to be set to "On Demand" for the scenario to be accessible as a tool in AI agents.

Integrating with OpenAI (ChatGPT)

1. Go to platform.openai.com and create an account if you don't have one.
2. Navigate to the dashboard and create a new chat.
3. In the chat interface, go to "Tools" and select "Create MCP Server."
4. Add the MCP URL copied from Make.com.
5. Label the server (e.g., "Make MCP Server").
6. Ensure there is no authentication required.
7. Click "Connect."
8. Update the tools list if needed to reflect the "On Demand" triggers in Make.com.

Integrating with Claude

1. Search for "make.com MCP server" on Google and find the official Make.com documentation.
2. Copy the code snippet provided for "Usage with Claude Desktop."
3. Open Claude Desktop and go to "Settings" -> "Developer" -> "Edit Config File."
4. Delete the existing content in the config.json file.
5. Paste the copied code snippet into the config.json file.
6. Replace the placeholder URL in the code with the MCP URL copied from Make.com, ensuring the URL is within the quotation marks and on a single line.
7. Close and reopen Claude for the changes to take effect.
8. Verify the integration by checking for the Make.com icon in Claude, which should display the available scenarios.

Configuring the Make.com Scenario

• Defining Input/Output Variables:
  • In the first module of the Make.com scenario (the MCP Server trigger), define the variables you want to extract from the AI agent's input (e.g., first name, last name, email, date).
  • Use the "input/output" button to add new variables.
  • Provide a description for each variable to give the AI agent more context.
  • Specify the data type for each variable (e.g., text, date).
• Returning a Value:
  • The last module in the scenario should return a message to the AI agent confirming the successful execution of the task (e.g., "New row added to Google Spreadsheet").

OCR Accuracy Considerations

• The video notes that OCR accuracy can vary between AI models (Claude vs. ChatGPT).
• In the invoice example, the pricing data extracted by Claude was inaccurate due to poor OCR performance.

MCP Client Modules: A Critical Assessment

• Make.com also released MCP client modules, which allow Make.com to interact with external AI agents.
• However, the video expresses a critical view of the current implementation, deeming it "kind of useless" due to limited functionality.
• Limitations:
  • Limited number of supported applications (around 15).
  • Lack of useful functionality within those applications (e.g., Appify not offering web scraping).
  • Redundancy with existing Make.com modules (e.g., MCP client for Asana providing the same functionality as the regular Asana module).

Conclusion

The video provides a detailed walkthrough of setting up and using Make.com's MCP functionality with ChatGPT and Claude. It highlights the benefits of MCP, such as tool discovery, merging methods, and natural language processing, while also offering a critical assessment of the current limitations of MCP client modules. The step-by-step instructions and practical examples make it easy for viewers to implement MCP in their own automation workflows. The video also emphasizes the importance of OCR accuracy and provides guidance on configuring Make.com scenarios to effectively extract and process data from AI agents.