Connect Agents to 100+ MCP Servers

Today’s top AI Highlights:

1. A developer-first open source autonomous AI agent framework

2. Connect AI agents, models, and IDEs to 250+ managed MCP servers

3. Mistral releases the “world's best document understanding API”

4. First general AI agent combining AI reasoning with autonomous computer actions

5. Chaos Coder generates 9 variations of web apps simultaneously

& so much more!

Read time: 3 mins

AI Tutorials

AI Agent Tutorial

Air quality has become a crucial health factor, especially in urban areas where pollution levels can significantly impact our daily lives. While many air quality monitoring tools exist, there's a gap when it comes to personalized health recommendations based on real-time air quality data.

In this tutorial, we'll walk you through building a multi-agent AQI Analysis App that gives personalized health recommendations based on real-time air quality data. This system will analyze current air conditions and provide tailored advice based on your health conditions and planned activities.

Tech stack:

• Firecrawl for web scraping

• Agno (formerly Phidata) to create and coordinate AI agents

• OpenAI GPT-4o as LLM

• Streamlit for interface

AI Workflow

This workflow combines Grok-3's image generation capabilities with Pika AI's video animation features to create stunning transformation videos that show the evolution from vintage to modern aesthetics. Perfect for photo restorations, concept visualizations, or creative storytelling.

We share hands-on tutorials like this 2-3 times a week, designed to help you stay ahead in the world of AI. If you're serious about leveling up your AI skills and staying ahead of the curve, subscribe now and be the first to access our latest tutorials.

Don’t forget to share this newsletter on your social channels and tag Unwind AI (X, LinkedIn, Threads, Facebook) to support us!

Latest Developments

Opensource Framework to Build, Manage and Run AI Agents 🏗️🤖🏢

SuperAGI is a developer-first, open-source framework for building and managing autonomous AI agents. The platform lets you build, manage, and run AI agents that can concurrently handle multiple tasks, integrate with a variety of tools, and continuously learn.

With features like refined token usage control and an expanse vector database support, SuperAGI gives you powerful tools for cost-effective and performant AI agent applications. The detailed level of control and customization available makes it worth exploring for any project involving autonomous agents.

Key Highlights:

1. Modular Architecture - SuperAGI is built on a modular architecture with a central executor, task queues, and separate components for models, tools, and knowledge management.

2. Agent Memory and Learning - It supports persistent agent memory storage. This enables agents to learn and adapt over time, going beyond simple task execution to become stateful, context-aware entities.

3. Granular Agent Control - A dedicated "Action Console" allows you to interact with agents in real time, providing input, granting permissions, and guiding execution. This is particularly useful for permission-based ("Restricted") agents.

4. Toolkit Ecosystem - Agents can connect to external services (Google Search, GitHub, etc.) via pre-built toolkits. You can also build custom toolkits.

5. Performance Monitoring - Built-in telemetry provides insights into agent performance (token usage, execution times, etc.). Combined with configurable constraints and iteration limits, this allows for fine-tuning and cost optimization.

6. Deployment Options - SuperAGI offers both a managed cloud service (for quick starts) and a local, Docker-based installation (for full control and customization, including GPU support).

Build Agents Connected to 100s of MCP Servers 🌐🔌

AI agents are only truly useful when they can interact with the real world, and Anthropic's Model Context Protocol (MCP) is quickly becoming the standard for making that happen.

Now, two key players in the agent space, Composio (integration platform for AI agents) and Agno (AI agent building framework), have announced major updates around MCP. Composio offers a drastically expanded library of managed MCP servers, while Agno has released tools to build MCP-powered agents with ease. These updates focus on providing broader access to applications, streamlined automation, and more flexible agent development.

Key Highlights:

Composio

1. Massive Integration Catalog - Composio now supports over 250 applications via fully managed MCP servers, including complex integrations like Salesforce and Hubspot, along with tools like Gmail and GitHub. Composio handles authentication automatically, removing a major development hurdle.

2. Near-Zero-Code Integration - Connecting to Composio's MCP servers can be done with minimal code. For example, integrating with Claude Desktop often involves a single npx command (provided by Composio) to install and configure the server connection.

3. Advanced Automation - Composio goes beyond basic data retrieval by offering 20,000+ API actions and triggers. You can build complete, automated workflows, and even convert any OpenAPI spec into an MCP server for custom integrations.

Agno

4. Lightweight Agent Framework - Provides a Python library for building MCP-connected agents with clear patterns for error handling and resource management.

5. Ready-to-Use Examples - Includes complete code examples for filesystem navigation, GitHub repo exploration, and more that you can adapt for your own agent projects.

Quick Bites

Mistral has launched its OCR API which they claim is "the world's best document understanding API" that comprehends images, text, tables, and equations with unprecedented accuracy. Mistral OCR processes documents as prompts and converts them into structured outputs for downstream function calls.

• Outperforms leading OCR models in benchmarks with 94.89% overall accuracy and excels particularly in math (94.29%) and tables (96.12%)

• Processes up to 2000 pages per minute on a single node, making it the fastest in its category

• Available now on la Plateforme at 1000 pages per dollar (~double the pages through batch inference)

• Selective self-hosting options for organizations handling sensitive or classified information

Chinese entrepreneur “Peak” Ji Yichao has released Manus AI, the first “general AI agent" that operates with its own dedicated computer, working asynchronously even when users are offline. It combines AI reasoning with autonomous actions, allowing it to independently handle files, browse websites, write code, access APIs, deploy websites — basically everything while maintaining memory of previous interactions and learning from user feedback. It’s like OpenAI’s Deep Research meets Operator agent.

Built as a multi-agent system powered by several distinct models, Manus claims leading performance on the GAIA benchmark. Currently in an invite-only preview, the team plans to open-source parts of the system later this year.

ChatGPT desktop app on Mac can now directly edit code within your IDEs like Xcode and VS Code, as per your natural language prompts. It can write code changes for you - showing what it plans to modify before applying them - or automatically implementing them while you work. Available to Plus, Pro, and Team users.

Tools of the Trade

1. Chaos Coder: Next.js application that generates 9 different variations of web applications simultaneously based on user prompts. It features real-time code previews, voice input support, and performance metrics.

2. LLM.txt Generator API by Firecrawl: Concatenates any websites into clean text files (llms.txt and llms-full.txt) specifically formatted for LLMs. It crawls the specified websites, extracts content, and organizes it into either concise summaries or complete text formats that can be directly fed into any LLM.

3. Chunkr: Open-source document parsing infrastructure to convert complex documents into LLMs and RAG-ready formats. It offers granular control over the parsing pipeline, enabling customized processing at the segment level (titles, tables, etc.) with options like OCR, VLM processing, and custom chunking strategies.

4. AgentFence: Open-source testing framework that detects vulnerabilities in AI agents. It automates adversarial testing, uncovers prompt injection attacks, detects secret leaks, and evaluates an AI model’s robustness against manipulation. Supports LangChain and OpenAI, more planned.

5. Awesome LLM Apps: Build awesome LLM apps with RAG, AI agents, and more to interact with data sources like GitHub, Gmail, PDFs, and YouTube videos, and automate complex work.

Hot Takes

1. Some developers are afraid of docker in a very unhealthy, career-blocking, you-won't-make-it way.
   You might not want to talk about it, but this is blocking you big time. ~
   Santiago
   
   

2. The inability of Meta’s Llama models maintain a lead in open weights, and LLMs overall, has been a bit surprising. They have the talent and compute that you expect could keep up with Anthropic, Xai, OpenAI and Google, let alone the top Chinese & French models, but they haven’t. ~
   Ethan Mollick
   
   

3. Friend just quit his job to start working on MCP servers ~
   anton

That’s all for today! See you tomorrow with more such AI-filled content.

Don’t forget to share this newsletter on your social channels and tag Unwind AI to support us!

Unwind AI - X | LinkedIn | Threads | Facebook

Awesome LLM Apps | Sponsor Us

PS: We curate this AI newsletter every day for FREE, your support is what keeps us going. If you find value in what you read, share it with at least one, two (or 20) of your friends 😉