Arcade.dev | The Actions Runtime

Anthropic just released Opus 5, should you be switching over from Fable 5? Short version, @ptdamiba says yes. Longer version, you'll have to watch the video. Have you tried Opus 5 yet? What's your initial take?

Two cities, two dinners this week. DC and Minneapolis brought out leaders from Wells Fargo, Cornerstone OnDemand, Optum, Medtronic, UnitedHealth Group, Lululemon, McKesson, Booz Allen, and more. We skipped the panel format and asked everyone to just talk about what AI looks like at their org right now. Two hours later we were still going. More coming: keep an eye out for your city!

Four new agent-optimized toolkits just shipped: Power BI, Postman, Fireflies, and Insightly, plus upgrades to Gmail, Excel, Outlook, and Linear. One of our customers was slowed down because Microsoft's MCP server for Power BI required a GUID just to ask "show me last quarter's revenue.” So we fixed it. arcade.dev/blog/new-toolk…

Session IDs are going away on July 28. Our founding engineer Sterling built Arcade's runtime and breaks down what actually changes and what to check before the release lands. Simple servers get simpler. Multi-user scale stops fighting you. More on the blog: arcade.dev/blog/session-i… All coverage: arcade.dev/mcp26 #MCP728

Last week we launched SkillBench. Three things 39,000 scored skills showed us: - 73% carry elevated safety risk and 7,034 were quarantined outright. - 32% of skills reach beyond what they declare, touching credentials or the network without saying so. - Among the 600 most popular skills, 62% are graded C or below. Check your scores: skillbench.arcade.dev. More on the blog: arcade.dev/blog/skillbenc…

July 28 is the largest MCP spec update to date. We’ve covered stateless and auth. Next up: extensions. Notice that no client is obligated to support any extension and authors can ship breaking changes on their own cadence. A tool that works in Claude might not work in Cursor. Read more for the full breakdown: arcade.dev/blog/mcp-exten…

TechCrunch needed someone to explain the July 28 MCP release. They called @nbarbettini. His verdict on today's setup: it "fights the load balancer instead of working with it." The good news is the new MCP release will be an unlock for businesses running agentic AI at scale. techcrunch.com/2026/07/20/ais… #MCP728

Deep dive on our blog: arcade.dev/blog/mcp-refre…

July 28 is the largest MCP spec update to date. Our engineer @WilsDawson contributed to this release, clarifying how MCP clients should handle refresh tokens. If it worked fine in the demo and then broke the morning after when you rolled it out to real users, refresh tokens might be your answer. Wils has been watching this break in production for a while. He fixed it. #MCP728

July 28 is the largest MCP spec update to date and we’re here to guide you through it. First up: stateless. The protocol is going stateless and session IDs are going away. If you run MCP servers at scale this is a breaking change that will force a rewrite. More coming before #MCP26 drops. Start with @nbarbettini's full post: arcade.dev/blog/mcp-going…

The new MCP spec drops July 28 and some things are going to break. If you are running MCP servers, you need to know what is coming. Never fear, @torresmateo is hosting a webinar the same day at 10AM Pacific. Register: arcade.dev/webinar/ #MCP26

Our CTO and co-founder @SamPartee spoke at #VBTransform2026 this week on why enterprise agents cap out at a few dozen tools and how Arcade is approaching a solution built around intent and tool quality that doesn't burn context budget or bypass governance.

MCP is graduating to much bigger things on July 28 when the new RC lands. Come party w/us and @TryArcade for talks and knowledge. "We unplug a live server mid-task and demo stateless in action" = live demos @ptdamiba will run the show. MCP728 gets real. luma.com/ugfl1mn2?tk=p8…

@TryArcade @torresmateo @avoguru Well, we kinda think you guys are doing pretty cool work over there on the agentic AI and MCP thangs and are lucky to have @torresmateo and @avoguru aboard.

Just like we established the benchmark for MCP quality with ToolBench, today we’re launching SkillBench, a quality benchmark for agent skills. Skills have grown to tens of thousands with no reliable way to judge quality before you adopt. SkillBench has already scored 39K skills including popular skills from Anthropic, Microsoft, Vercel and more, each scored across six dimensions that matter most to enterprise businesses. Check skills and submit your own at skillbench.arcade.dev/browse

Three major AI labs shipped enterprise-focused releases within 48 hours of each other last week: GPT-5.6, Grok 4.5, and Muse Spark 1.1. Our CEO and co-founder @TheMostlyGreat on what enterprises still need before any of these deployments actually work in production: arcade.dev/blog/chatgpt-w…

Two members of the Arcade team were selected as AAIF Ambassadors. Congrats to @torresmateo and @avoguru on joining the inaugural cohort of the @AgenticAIFDN Ambassador Program, 140 ambassadors across 41 countries 👏 More about the program: #list" target="_blank" rel="nofollow noopener">aaif.io/ambassadors/#l…

Our CTO @SamPartee is speaking at #VBTransform2026 on 7/14. He’ll unveil a new Arcade innovation that will shift what’s possible around agents and tool discovery. That’s all the details we can share now, you’ll have to catch the presentation… Register: luma.com/92nbdnnx

Arcade sits between every agent and the systems it acts on. When something breaks, does everything go down with it? No. Most failures come from 3rd party services, not the runtime. And if something goes wrong, every action is scoped to one user so the blast radius stays small. Read the full breakdown: arcade.dev/blog/is-arcade…