- unwind ai
- Posts
- Research Paper to AI Agents with MCP
Research Paper to AI Agents with MCP
+ Google's free 5-Day AI Agents Intensive course, Cloudflare open-sourced VibeSDK
Shubham Saboo & Gargi Gupta
September 25, 2025
Today’s top AI Highlights:
& so much more!
Read time: 3 mins
AI Tutorial
Learn OpenAI Agents SDK from zero to production-ready!
We have created a comprehensive crash course that takes you through 11 hands-on tutorials covering everything from basic agent creation to advanced multi-agent workflows using OpenAI Agents SDK.
What you'll learn and build:
Starter agents with structured outputs using Pydantic
Tool-integrated agents with custom functions and built-in capabilities
Multi-agent systems with handoffs and delegation
Production-ready agents with tracing, guardrails, and sessions
Voice agents with real-time conversation capabilities
Each tutorial includes working code, interactive web interfaces, and real-world examples.
The course covers the complete agent development lifecycle: orchestration, tool integration, memory management, and deployment strategies.
Everything is 100% open-source.
We share hands-on tutorials like this every 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.
Latest Developments
Picture this: you find a research paper with exactly the method you need, but after downloading the code, you're stuck wrestling with broken dependencies, cryptic documentation, and parameter configurations that make zero sense.
Stanford researchers just dropped Paper2Agent, an open-source framework that converts static research papers into interactive AI agents you can actually talk to and use. We usually don't post academic papers, but this addresses something almost every developer must have experienced firsthand.
Paper2Agent automatically analyzes papers and their codebases, wraps the core methods into standardized tools using MCP, and creates agents that can reproduce results, apply methods to new data, and even guide you through complex workflows via natural language.
Key Highlights:
Multi-Agent Architecture - Uses a team of agents for environment setup, code extraction, tool implementation, and testing validation, with each component working autonomously to ensure reliable conversion from paper to working agent.
Tool Generation - Scans research codebases and automatically creates 6-22 reusable tools per paper, covering everything from data preprocessing to visualization, with built-in testing to ensure 100% accuracy against original results.
MCP Integration - Built on MCP for seamless integration with any AI agent or LLM, allowing researchers to combine tools from multiple papers in a single conversation without additional setup.
Production Ready - Generated agents are hosted remotely on platforms like Hugging Face Spaces, eliminating local dependency issues and providing instant access to validated, reproducible implementations.
AI coding agents generate 1000s of lines of code daily, but there’s a fundamental problem - they are not able to see what the code they generate actually does when it runs in the browser. They're effectively programming with a blindfold on.
Google has open-sourced Chrome DevTools MCP server that enables AI agents to see exactly what happens when your code hits the browser.
It connects the agent directly to a live Chrome instance, where it can inspect DOM elements, monitor network requests, and analyze performance metrics like a human developer would. Whether you're using Claude, Cursor, or Copilot, your AI can now verify fixes as they're applied, diagnose CORS errors by examining actual network traffic, simulate user interactions to reproduce bugs, and run automated performance audits with actionable insights.
Key Highlights:
26 specialized tools - The server provides input automation, navigation control, performance tracing, network analysis, and debugging capabilities through a single MCP integration.
Automated user simulation - Tools for clicking elements, filling forms, scrolling pages, and waiting for specific text to appear enable comprehensive user flow testing by AI agents.
Performance optimization tools - Built-in CPU throttling, network condition emulation, and detailed trace analysis help AI identify and fix performance issues with data-driven insights.
Deployment - Configurable security options including sandboxing support, custom Chrome executable paths, and temporary isolated user data directories for secure automation.
Built for Managers, Not Engineers
AI isn’t just for developers. The AI Report gives business leaders daily, practical insights you can apply to ops, sales, marketing, and strategy.
No tech jargon. No wasted time. Just actionable tools to help you lead smarter.
Start where it counts.
Quick Bites
Deploy your own vibe coding app, end-to-end, with just one click
Cloudflare has open-sourced VibeSDK, an AI-powered vibe coding platform you can deploy in just one click. It comes with secure sandboxes for running AI-generated code, integration with LLMs for building and debugging apps, built-in templates, and one-click export to GitHub or Cloudflare accounts. Teams and SaaS products can embed this to let users spin up applications instantly, with full control over hosting, privacy, and scaling.
Free 5-Day AI Agents Intensive course by Google and Kaggle
After their GenAI course hit 420,000 registrations last year, Google's back with a 5-Day AI Agents Intensive running November 10-14. The course architecture covers core components - models, tools, orchestration, memory, and evaluation - with hands-on experience in Google's MCP and A2A protocols. Daily assignments include research papers, NotebookLM-generated podcasts, and live AMAs with Google's ML team, culminating in a portfolio-worthy capstone project.
Claude is now available in Microsoft 365 Copilot
Claude Sonnet 4 and Opus 4.1 are now available within Microsoft 365 Copilot, starting with the Researcher agent and Copilot Studio for enterprise customers. The integration gives organizations model choice for complex reasoning tasks, from building go-to-market strategies to creating custom enterprise agents. It's rolling out today through Microsoft's Frontier Program to opted-in organizations.
Lovable can now turn files into apps and websites
Lovable now accepts direct file uploads and converts them into functional apps. Drop in a CSV for an interactive dashboard, a PDF resume for a portfolio site, or a PowerPoint deck for a custom presentation. The platform automatically parses different file formats and generates appropriate web applications, handling everything from data visualization to media embedding.
Gemini CLI now seamlessly integrates with FastMCP
Google's Gemini CLI now integrates directly with FastMCP, a widely-used Python library for building MCP servers. This integration will let you install local STDIO transport MCP servers built with FastMCP, with a single command, automatically handling configuration and dependency management. This makes custom tools and prompts feel native within the terminal environment, significantly expanding what Gemini CLI can accomplish.
GPT-5-Codex is now available via API
OpenAI has released GPT-5-Codex in the Responses API. If you use the Codex CLI via API key, you can now also use GPT-5-Codex. With a 400,000 tokens context window, the API costs $1.25 (input) and $10 (output) per million tokens.
Data Commons MCP server for easy public data access to AI agents
Google has released a Data Commons MCP Server that lets AI agents directly access its vast repository of public datasets via MCP. This enables AI agents to ground their responses on real data, while handling the full spectrum of data queries from exploratory searches to analytical comparisons, and generating complete reports.
Tools of the Trade
Inferencer - MacOS app that runs AI models locally while giving you granular control over the inference process, including real-time token probability inspection and entropy visualization. You can manipulate token selection during generation and prefill prompts to direct model responses, all while keeping processing entirely on-device.
Snapdeck - Generates professional presentations using an orchestration layer that routes tasks across multiple language models and APIs to create fully editable decks. Give it your requirements in plain English or pull content from websites/Notion, then modify layouts, visuals, and content through direct manipulation or simple prompts.
Mixboard - Google Labs' experimental AI-powered visual concepting board that lets you explore and refine ideas through an open canvas interface. With a simple prompt, it can generate and edit full moodboards, visual concepts, and help you brainstorm for projects ranging from home decor to product ideas.
Awesome LLM Apps - A curated collection of LLM apps with RAG, AI Agents, multi-agent teams, MCP, voice agents, and more. The apps use models from OpenAI, Anthropic, Google, and open-source models like DeepSeek, Qwen, and Llama that you can run locally on your computer.
(Now accepting GitHub sponsorships)
Hot Takes
Everything that AI hypermarketers claimed was going to be dead by now is very much alive:
• CS degree — do it if you can
• Coding — learn it, period
• RAG & prompt engineering — roughly all you need
• Evals — no pain, no gain
• Wrapper startup — pay close attention to your users ~
elvisIt's interesting how "better at code" has become the defining goal of almost every AI lab over the last twelve months
I think Claude Code getting a bunch of people onto $200/month plans proved that code is one of the most economically valuable applications of this technology ~
Simon Willison
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!
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 😉
Reply