Arcade.dev

Just published a video demonstrating a real problem: when your AI agent needs to access user services, complexity explodes. We took a simple Telegram bot that schedules Google Calendar events and showed the reality of what it takes to make it work: - OAuth 2.0 implementation - Token storage/refresh logic - API endpoint management - Error handling across multiple services All this for a SINGLE integration on a SINGLE platform.Then we rebuilt it with Arcade.dev. Not only did the auth code shrink to 9 lines, but I completely eliminated: - The Express server for OAuth callbacks - All Google API client code - Token management complexity Most importantly: extending this to Slack, WhatsApp, or Discord now only requires handling the new platform's messaging interface. The core logic with Groq LLM and the Telegram bot architecture remain unchanged. Try the code yourself - link in comments.

WALL STREET JOURNAL CONFIRMS IT: Authorization is the final boss battle for AI agents. The race to build smarter AI models? That was the easy part. The WSJ just called out what we've known since day one: AI agents are useless if they can't securely act on your behalf. No access to APIs, no booking flights, no sending emails from your accounts. Our CEO Alex Salazar laid it out clearly in the article: "Tool-calling agents are the emerging phase of agent AI development." Translation: LLMs without action are just fancy chatbots. The article spotlights how we're helping Shortwave's email agent connect to Notion – letting their AI actually DO something instead of just talking about it. And that's just the beginning. This isn't incremental progress. The WSJ compares this shift to the 2008 app store revolution that fundamentally changed how humans interact with technology. We've been quietly building this future while others were still figuring out prompt engineering. Now the world's catching up. Read the full story: https://lnkd.in/gZwy9fY8 What systems would you unlock first if your AI could securely access them as you? (Spoiler: we probably already built that connector)

Well that escalated quickly! MCP just had its biggest week yet:  🚀 LangChain dropped their Open Agent Platform powered by Arcade.dev - using Arcade's integration library and MCP support to securely connect agents with external services  🏢 Visual Studio Code is in the game too - Harald Kirschner published a deep dive into their MCP integration journey  👁️ Even PwC is jumping on the MCP train with their agent OS announcement  💰Sequoia Capital's Konstantine Buhler took MCP mainstream on Bloomberg, explaining how this innovation is pivotal for cross-platform communication and boosting business efficiency The rapid adoption across frameworks, developer tools, and enterprise platforms signals MCP's evolution from experimental protocol to essential infrastructure. What's your take on the MCP explosion? #MCP #AIAgents #ModelContextProtocol

"Your agent keeps failing because it can't login as you." Our team at Arcade.dev is showing the escape hatch: production-ready auth that lets LLMs safely execute with user permissions across Gmail, Slack, and Salesforce. Come see our engineers demonstrate how we solved today's biggest agent blocker with zero browser hacks or janky workarounds. Thanks to the LangChain crew for building this community! If you're here, find our crew in red for live demos and the most technically honest conversation about agent auth you'll have all week.

"Auth is where AI agents go to die." Sound familiar? Yesterday's panel with Nate Barbettini and William Dawson from our team plus Den Delimarsky (Microsoft) and Aaron Parecki (Okta) hit this problem head-on. We've all been there: Your AI agent is CRUSHING IT locally with your personal token... then security review happens. Dead in the water. The hard truth? The gap between "cool demo" and "enterprise-ready" is almost entirely about authentication: - Your personal tokens won't scale beyond your laptop - AI agents need to access 10+ services simultaneously - Each service has completely different auth requirements - Security teams (rightfully) freak out about access control This is exactly why we built Arcade.dev – we were tired of watching brilliant AI projects die during security review. If you're building agents that need to DO things beyond just chat, watch this full discussion (link in comments). The MCP community is solving real problems that every dev team will face as AI moves to production. Who else is hitting auth walls with their AI projects? Drop your stories below! 

Excited to have Arcade.dev join us at Interrupt! They give agents the power to take real actions for your users. Swing by on May 14th to learn how Arcade’s auth and tool APIs connect your agent to Gmail, Slack, HubSpot, and your favorite MCP servers in just a few lines of code. 🔗 interrupt.langchain.com #LangChain #Interrupt2025 🔗 https://arcade.dev

So you want your MCP agents to escape the local sandbox? MCP crushes it locally.. But the moment you deploy to the cloud, things get... interesting. Proxying stdio over HTTP? Brokering OAuth for downstream resource servers? That's when even the bravest developers start questioning their life choices. The new HTTP transport is a massive win, but authorization in the cloud is hard. Who's making this call? Are they allowed to? That's the question keeping cloud security teams up at night. Our team has been working with security experts from Microsoft, Google, Okta, AWS and more to build a proper auth spec for MCP. See our take on the future of MCP – and what cloud agents need to thrive: https://lnkd.in/gC7pxMsD

What if LLMs could *do* things?  I’m talking about more than an MCP server on your laptop wrapping an API, which is a good start, but what about production-grade tool-calling that can take complex actions with intentionality? Agents want “reply to my latest email thread with Sally, and include this attachment” and “update the Salesforce contact and close the opportunity”, not “read the last email in thread 123abc”.  Each of those intentions wraps a few different APIs, needs retries and internal decision logic, a strong security posture… and it has to work every time with no hallucinations. That’s why I joined Arcade.dev as the Head of Engineering. I’m continuing on the path I started with Grouparoo and then at Airbyte - to make the best developer tools to deal with data. We are building the development kits, tools, runtime, and registry that are needed break LLMs out of the chat box and make productive agents at-scale 🤖.

The AI agent ecosystem is accelerating at breakneck speed! Here's what happened just this week: 🤝 Google released their A2A protocol - the universal translator for agent systems that enables different frameworks like CrewAI and LangGraph to seamlessly collaborate 🛡️ The MCP spec received a major security enhancement with input from top auth experts at Microsoft, Arcade, Okta, Stytch, Anthropic and more ⚡ Arcade.dev demonstrated building a complete multi-agent system in under an hour using OpenAI's Agents SDK - a task that previously required days and specialized expertise 🚀 xAI's Grok v3 models are now available via API, offering developers a less filtered, more creative option in their AI toolkit ☁️ And while not from this week (but still worth highlighting) - CloudFlare recently showed how to deploy MCP servers globally in minutes using their Workers platform, complete with OAuth and proper auth flows The pace of innovation in agent technology is stunning - we're witnessing the infrastructure for truly collaborative AI systems being built right before our eyes. What development are you most excited about? #AIAgents #AITools #AgentSystems

Just dropped: Build your own AI Gmail Agent in less time than it takes to argue about tabs vs. spaces! 🚀 Ever wanted to ask your inbox "Who's been emailing me about that deadline?" and get an actual answer instead of cold, digital silence? We built a sample app that shows you EXACTLY how to make that happen with Python, Streamlit, and OpenAI. The really cool part? We solved that nightmare OAuth flow so your users can securely connect their Gmail accounts without you having to sacrifice your sanity to the authentication gods. One-click login that works? What is this sorcery?! (It's Arcade.dev) Complete with: • ALL the code you need on GitHub • Mateo Torres's Step-by-step video walkthrough (only 40 min!) • Data structure blueprints for Gmail API wrangling • Prompt engineering patterns that actually work Build it this weekend, put it in your portfolio on Monday, and watch your colleagues slowly realize you've become too powerful to be left alone with an API key. Watch the full tutorial: https://lnkd.in/e8pXz79Q Grab the repo: https://lnkd.in/eYj__fFe Now go forth and teach email who's boss! 💪 #AIEngineering #AIAgents #WeekendProject