A: Hey a.quartarone,

Let's break down your questions:

Using Natural Language for Workflows: You don't currently type "when Gmail then Airtable" and have AgenticFlow automatically build a visual workflow diagram like you might see in Zapier or Make. However, we offer something designed to be even more intuitive: using an Agent with MCP.

You connect the necessary tools (like Gmail and Airtable) to your workspace once using our MCP integrations (Model Context Protocol).

Then, you instruct your Agent in plain English: "When a new email arrives with 'Invoice' in the subject, extract the customer name and amount, and add it as a new record in my Airtable base."

The Agent understands your request and uses the connected MCP tools (Gmail, Airtable) behind the scenes to perform the task. You don't manually wire up the nodes; you delegate the task using natural language.

You can see this agent-based approach in action here: https://youtu.be/dsugnFmHUMM?si=Y5qIcXbcAJuOhV5g

Confirming Text-Based Instruction: Yes, you can achieve the outcome of automating tasks like "Gmail to Airtable" simply by writing text instructions to an Agent that has the required MCP tools connected. It figures out the steps. (We are exploring a future feature to convert agent conversations into visual workflows too - vote for it on the roadmap if interested: https://agenticflow.featurebase.app/).

Using MCP Apps vs. APIs: The goal of MCP is to make using the 2500+ integrations simpler than dealing with raw APIs directly, much like Zapier simplifies connections.

For most MCPs (like Gmail, Google Sheets, Airtable, Flodesk, etc. - see the list: https://agenticflow.ai/mcp), you connect your account once using a standard method (like OAuth login or pasting an API key into the MCP setup page).

After that one-time setup, the tool is available for your agents to use based on your natural language instructions. You don't need to handle API calls, headers, or JSON parsing yourself for these integrated apps.

Think of it as giving your agent a ready-to-use tool from the toolbox, rather than needing to build the tool yourself from raw API parts every time like you might with a generic HTTP node.

So, while it's not text-to-visual-workflow, it is text-to-automated-action using pre-connected tools, aiming for simplicity closer to Zapier than raw API coding.