- unwind ai
- Posts
- Build and Deploy Remote MCP Servers in Minutes
Build and Deploy Remote MCP Servers in Minutes
PLUS: New Gemini 2.5 Pro with coding improvements, AI dev agent earned $88K on SWE bench
Today’s top AI Highlights:
Build and deploy Remote MCP servers with OAuth and state in Cloudflare
This autonomous dev agent just earned $88K on a SWE benchmark
OpenAI is buying Windsurf; Anthropic and Apple are building a vibe coding platform
New Gemini 2.5 Pro comes with major front-end development skills
Super-fast computer-use general-purpose AI agent free to try
& so much more!
Read time: 3 mins
AI Tutorial
Integrating travel services as a developer often means wrestling with a patchwork of inconsistent APIs. Each API—whether for maps, weather, bookings, or calendars—brings its own implementation challenges, authentication systems, and maintenance burdens. The travel industry's fragmented tech landscape creates unnecessary complexity that distracts from building great user experiences.
In this tutorial, we’ll build a multi-agent AI travel planner using MCP servers as universal connectors. By using MCP as a standardized layer, we can focus on creating intelligent agent behaviors rather than getting bogged down in API-specific quirks. Our application will orchestrate specialized AI agents that handle different aspects of travel planning while using external services through the MCP. We'll be using the Agno framework to create and orchestrate our team of specialized AI agents.
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
Many folks probably missed what Anthropic released with the new Integrations in Claude announcement: full remote MCP support. Until now, we could only connect MCP servers locally to Claude Desktop. Now, you can simply paste a remote MCP server URL into Claude and use it directly on the web or desktop. No more local setup, no Docker, no CLI work. Just share a URL, and Claude connects.
If you’re looking to build and host your own MCP server, Cloudflare just made it dead simple. Their latest blog outlines how to deploy remote MCP servers using Workers, with built-in OAuth support, persistent sessions via Durable Objects, and easy integration into Claude and other tools using adapters. You can spin up a fully working server, plug in tools, and expose them to MCP clients in under 30 minutes. It handles transport, tokens, and state management out of the box.
Key Highlights:
Remote MCP + OAuth - Cloudflare’s workers-oauth-provider lets your MCP server act as an OAuth provider with just a few lines of code. You can use GitHub, Google, or any provider for login, with scoped permissions for tools.
Durable Objects for stateful sessions - Remote MCP servers on Cloudflare can maintain their own state. Each MCP client session gets a dedicated Durable Object, so your server can keep session-specific state. This would be great for applications like counters, shopping carts, persistent memory, or custom flows.
Streamable HTTP transport - Their new McpAgent class built into the Cloudflare agents SDK abstracts the transport layer, and will work with the newer spec revisions (streamable HTTP replacing SSE) without needing you to rewrite anything.
Test and use - Test your remote MCP server immediately with Cloudflare’s AI Playground chat interface or use the mcp-remote adapter to connect with MCP clients that don't yet natively support remote connections. This lets you preview the full experience today, without waiting for client apps to catch up.
Free AI Resource
Unlock powerful workflows from two CEOs who are reinventing the future of work with AI.
In this free resource, you'll discover how forward-thinking leaders are using AI to streamline meetings, cut busywork, and automate decisions.
Get actionable insights that you can implement in your own remote teams today.
Cosine’s fully autonomous software engineering agent, Genie 2 is out of beta and available for everyone to use. Built on a fine-tuned GPT 4.1 model, Genie combines strong reasoning capabilities with a native execution environment to test and run code instantly. Whether working independently or collaboratively, this AI engineer integrates with tools like GitHub, Slack, Jira, and Linear, making it a versatile addition to any development team.
Genie 2 achieved a record-breaking performance on the SWE-Lancer benchmark, earning $88K with a 49% resolution rate, outperforming most frontier models at a fraction of the cost. The best part? No need to manage compute resources or pay for "compute credits" - Genie handles everything behind the scenes.
Key Highlights:
Native Execution Environment - Genie 2 runs code directly in the platform without relying on external CI pipelines. It can compile code, run unit tests, and validate solutions in a sandboxed environment, providing instant feedback that's tightly integrated into its reasoning loop. All this happens invisibly to you - no need to provision machines or manage compute resources.
Flexible Working Modes - Use Genie asynchronously by assigning tasks while you focus elsewhere, or collaborate in real-time as it makes changes while you provide guidance. Genie takes the lead in making changes while you serve as the copilot.
Tool Integration - Genie works seamlessly with the tools you already use daily. It can pick up tasks from Jira or Linear, post updates in Slack channels, and commit code to GitHub, effectively becoming a virtual team member that communicates across the same platforms as your human team.
Purpose-Built Development Environment - Rather than squeezing AI capabilities into existing IDEs, Cosine built a lightweight, browser-based environment specifically for Genie. This helps with quicker iteration, faster performance, and workflows without the constraints of traditional IDEs.
Quick Bites
Just 15 days before the Google I/O 2025, the team has released a new version of Gemini 2.5 Pro. The updated model improves coding performance across the board, especially for frontend development, UI styling, and agentic workflows. Developers can already use it via the Gemini API and Vertex AI, with no changes needed if they were using the previous 2.5 Pro version.
Now ranked #1 on the WebDev Arena leaderboard, outperforming Claude 3.7 Sonnet, for building visually appealing and functional web apps.
Streamlines feature development by automatically handling design elements like colors, fonts, padding, margins, and borders when generating new UI components.
Gemini 2.5 Pro comes with state-of-the-art video understanding that you can combine with coding for innovative workflows like creating interactive learning apps from YouTube videos.
OpenAI is reportedly buying Windsurf, the fast-growing agentic IDE, in a $3B deal, marking one of its largest acquisitions yet, which could be announced as early as today. The acquisition gives OpenAI a stronger foothold in the developer tools space with Windsurf bringing its 800,000 developer users and 1,000 enterprise customers into the fold, to help them understand how technical professionals use competing LLMs.
On the other side, Anthropic is partnering with Apple to build a vibe coding platform that’ll use Anthropic Claude 3.7 Sonnet and Apple’s programming platform Xcode. Nothing is official yet, though.
Tools of the Trade
Scout by Scrapybara: A super-fast general-purpose AI agent with its own virtual Ubuntu computer that can browse the web, run terminal commands, write code, and handle files. You give it a prompt and optional files, and it works independently—whether for quick tasks or long-running jobs
CodeCafé: A real-time, browser-based code editor for collaborative applications programming, teaching, and web development. You can see your HTML, CSS, and JavaScript changes render instantly in an integrated Web View.
Well: MCP-compatible agent that automates supplier invoice collection across portals, emails, WhatsApp, and other channels using AI-based workflows. This agent auto-detects invoices as you browse, batch-collects them during tax season, and adapts to UI changes without manual fixes.
Awesome LLM Apps: Build awesome LLM apps with RAG, AI agents, MCP, and more to interact with data sources like GitHub, Gmail, PDFs, and YouTube videos, and automate complex work.
Hot Takes
My humble suggestion to OpenAI:
Their most advanced model should be a member of all their boards.
The context should be: do the actions decided by the board match the mandate and principles laid out.
If you can’t convince the AI you can’t do it. ~
Chinese startups are going to great lengths to not be seen as Chinese.
Manus competitor Genspark, for example, has:
— No employees on LinkedIn. Reports say they're in China
— Founders worked in Beijing at Baidu for most of career
— Press explicitly describes them as "Palo Alto based"
— Seed backer Lanchi used to be China's 4th largest AI investor
Good product, at $22M ARR, and I imagine they do this just to be attractive to US investors or not get banned in the US. ~
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