A: Hey Never57,

Thanks for the great questions and for pointing out Agno! That's actually the first time I'm hearing about them, so appreciate you sharing their GitHub.

1. AgenticFlow vs. Agno.com:

From a quick look at their GitHub, Agno looks like a really interesting, powerful, and developer-focused Python library for building AI agents, especially with features like reasoning tools, multimodal capabilities, and a strong emphasis on performance for instantiating many agents. It's geared towards developers writing code.

AgenticFlow is fundamentally different in its approach: We are a no-code/low-code platform. Our goal is to make building sophisticated AI agents and workflows accessible to everyone, including non-technical users, marketers, business operators, etc., through a visual interface, pre-built templates, and natural language instructions.

While Agno provides the building blocks for developers, AgenticFlow provides the ready-to-use car with an intuitive dashboard. We handle the infrastructure, the UI, and simplify connections through our MCP layer.

2. MCP Server Costs & Self-Hosted:

Yes, calling an MCP server action does cost AgenticFlow credits. Each step in an agent's process or a workflow run has a small fixed platform credit cost (3-4 credits depending on your tier).

If the MCP action itself involves an LLM call (e.g., the MCP server internally calls an AI to process data before returning it to AgenticFlow), and if that LLM is one of AgenticFlow's built-in economical models, there would also be a variable credit cost for that token usage.

If you BYOK to the MCP server provider (e.g., if your ActivePieces, which hosts the MCP, is using your OpenAI key), then the LLM token cost is billed by that provider to you.

Self-Hosted MCP: Even with a self-hosted MCP, the AgenticFlow platform still incurs a cost to initiate the request, manage the connection, wait for the response, and process the result. So, the small fixed platform credit cost per step still applies. The agent also needs to use some tokens (and thus variable credits if using built-in LLMs) to "think" about what to send to your MCP and how to process the response.

3. Current Limitations (Beyond Multi-Agent):

While our 2500+ MCPs offer broad connectivity, we don't have native "Computer UI automation" (RPA-like clicking on a desktop).

Directly scraping highly protected sites (like social media feeds behind logins or complex JavaScript sites that actively block bots) can be challenging with our basic built-in scraper; for those, we recommend robust external services like Apify via its MCP.

Essentially, if a task requires very deep, code-level manipulation of a specific software that doesn't have an API, or involves overcoming sophisticated anti-bot measures on the web without a specialized scraping tool, AgenticFlow might struggle without a suitable MCP or API to connect to.

It's always best to test a specific complex use case with our free trial or a lower tier first.

4. Future Limitations (Fully Matured Product):
* Our vision is indeed for AI agent teams to handle almost any repetitive digital task a human can do on a laptop.

However, the ultimate capability will always depend on the advancements in underlying LLM reasoning, the robustness of APIs/MCPs for various tools, and the evolution of the entire AI ecosystem. We aim to be at the forefront of orchestrating these, but some tasks might always remain too nuanced or require human-level general intelligence that AI hasn't yet achieved.

Thanks for making me aware of Agno – always interesting to see other approaches in the agent space!
— Sean