MCP Tools to Make AI Agents: Rebuilding a CRM Assistant in Make AI Agents

In this hands-on tutorial, Vicente puts Make's new MCP server integration for AI Agents to the test, connecting his own CRM and automation tools live to see how far he can push it. After building a Telegram-based assistant that can query, reason over, and even talk back about his lead data, Vicente shows you exactly what works, what breaks, and how to fix it.

🔍 Inside the Build:

* The Big Idea: Learn why connecting a full MCP server to your Make AI Agent beats giving it individual tool modules one by one — especially when your action list starts getting crowded.

* Agent vs. Pipeline: Vicente breaks down when you actually need an AI Agent (research, personalized outreach, judgment calls) versus when a simple scenario will do the job faster.

* Live Troubleshooting: A real "it's not working, let's find out why" moment — Vicente hits a wall with the Airtable MCP, diagnoses the issue via direct agent chat, and pivots to HubSpot's official MCP instead.

* Bonus Build: Watch Vicente extend the assistant with a voice-note tool, chaining an agent response into ElevenLabs through a sub-scenario — including the fix needed when binary audio files trip up the agent.

* Real Results: See the assistant correctly summarize CRM data and respond with an actual voice memo in Telegram, end to end.

Vicente's mission is to show that MCP isn't just a new module — it's a way to give your Make AI Agent real autonomy over an entire platform. This video is a perfect blueprint for builders who want their agents to do more than follow fixed steps.

🛠️ Featured Stack:

* Make AI Agents (Core reasoning engine)

* MCP Servers (Airtable, HubSpot — tool access layer)

* Telegram (Chat interface & trigger)

* ElevenLabs (Text-to-speech for voice note replies)

* Google Drive (File hosting for agent-readable links)

💡 Pro-Tip from Vicente: "Always give your agent a reference point for time." Vicente highlights that AI Agents don't inherently know what "today" is — so passing the current date/time as an input is essential any time you're asking the agent to filter or reason over time-based data. 🔗 Make


https://www.youtube.com/watch?v=0qXLno7-GVk&utm_source=dlvr.it&utm_medium=blogger