Arcade.dev

The buyer is becoming the agent. That was the opening observation at our dinner this week in SF, hosted with Tavily during LangChain #Interrupt26. The room included leaders from Uber, Coinbase, Databricks, Oracle, Samsung, FedEx, and more. From there, we got into why most agent deployments stall. Fast prototypes become unmaintainable codebases before anyone notices. The table also reflected on MCP: what is actually working, what is still hype, and where it goes from here. The conversation closed on skills, specifically the security and governance of skills as agents take on more. Nobody has a clean answer yet. Tool quality is part of that answer. See how yours scores at toolbench.arcade.dev 🤝

Every AI agent that stores OAuth tokens server-side is building a token vault. The more integrations it supports, the more valuable that vault becomes to an attacker. The Vercel breach is the clearest example yet of what happens when that vault gets breached. One intrusion, tokens from every user who had ever connected the app, all ready to use at the provider level. The fix is architectural: tokens scoped per-tool, kept out of the LLM, auditable at every step, and for enterprises that need it, self-hosted so the blast radius stays within a single tenant. Mateo Torres wrote the full breakdown ⬇️

Today at the LangChain #Interrupt26 keynote, Harrison Chase featured Arcade as an MCP gateway partner for LangSmith Deployment. Arcade and LangChain set the benchmark for partnerships, pushing the industry forward on open standards, neutrality, and scale. Arcade's 7,500+ agent-optimized tools are now available in LangSmith Fleet through a single secure gateway, giving agents per-user authorization and reliable tool access across Salesforce, Asana, Slack, and more. Indispensable combo for any agent architecture 🤝 If you're at the conference, come say hi!

In today's webinar I'll show how to use MCP Apps with Arcade to return rich UI components to the MCP client. 📺 Join me live on Arcade.dev's YouTube channel at 13:30 PST. #AIAgents #MCP

Governance is the least exciting part of agents. It is also one of the most important. On May 13th, Arcade and Tavily are hosting a private dinner in San Francisco for senior AI leaders who know that solving governance is what unlocks the real potential of agents: how do you build the control plane that lets them move from proof of concept to production safely and at scale? If you are in town for LangChain #Interrupt26, come join us. Approval required. Apply at the link in comments ⬇️

Arcade sponsored a CIO and CISO Think Tank in Dallas with leaders from some of the largest enterprises in the country. The conversation kept circling back to the same tension: the pace of AI is moving faster than most organizations can govern it. Everyone in that room is trying to figure out how to give their teams the confidence to move while keeping visibility into what agents are actually doing. The format made a difference. One conversation around the table for most of the evening, just leaders sharing openly about how to close that gap, no sales pitches. Thank you to the organizers for having us!

Today we are open-sourcing Agent Library, a local-first memory layer for AI agents. Most agent memory today lives in someone else's product, in their format, behind their account. When you want to switch models, switch harnesses, or just take your context with you, you often can't. The memory was never really yours. Our CTO Sam Partee built Agent Library to change that. Agent Library is a local SQLite file with hybrid semantic and keyword search. It ingests text, code, PDFs, and images, speaks MCP natively, and runs entirely on your own machine. The most important design decision: from an agent's point of view, the library is read-mostly. Expose only the read tools and the agent can search your knowledge base all day. It cannot edit it. Read the full blog from Mateo Torres 👇

The thin line between magic and tragic: getting MCPs right. That is the name of the panel our CEO Alex Salazar is moderating tomorrow at the AI Agent Conference in NYC. 4:20 PM, Trianon Ballroom. Joining him: Prerit Munjal, Senior Technical Product Manager at Groupon, Ralph Bird, Principal Machine Learning Engineer at PagerDuty, and Ze'ev Klapow, Principal Software Engineer at HubSpot. Practitioners who have moved past experimentation and into production. Come join us!