Arcade.dev

Building agents is easier than ever with LangSmith Agent Builder. We built some Arcade-powered templates for the new Agent Builder Template Library. Giving you fully working agents with the authenticated tools they actually need to do real work: 💪Sales meeting prep 💪Lead capture & routing 💪Contract renewals 💪Customer health monitoring 💪Support ticket escalation No more skipping production day. These agents are ready to lift. Grab the templates and ship your agent: https://lnkd.in/gQt-bUsQ LangChain

Single-user demos? Light cardio. Multi-user production with secure user auth? That's the weight nobody wants to touch. You don't need a 6-month auth workout to ship multi-user agents. Arcade.dev is the only MCP runtime built for agent authorization—handling OAuth, token lifecycles, and granular permissions automatically. Your agents get controlled user-specific access. Your security team gets peace of mind. Stop skipping production day. Try for free at Arcade.dev

New year. Same lie: “Production-ready agent.” Cool. Now ask it to do literally anything. *Immediate sweating* We made 7 Minute Apps to call out why agents keep dying the moment action is involved. Stop building chatbots. Start shipping agents. 👉 Watch 7 Minute Apps: https://lnkd.in/g4yb6f8S

End-of-year reviews are a good time to be honest about what’s blocking AI agents from shipping. In our experience, it’s not the model. It’s what happens after the prompt — when agents need to take real actions for real users. If you don’t have a runtime that can enforce user-specific authorization, centrally govern MCP tools, and give security full visibility, the agent stays a demo. This is why production agents need a real MCP runtime. Need an MCP Runtime in 2026? Try Arcade.dev

What if your agent could actually run your workflow — not just talk about it? Today, we’re launching Arcade Gateways: a way for MCP clients to load agent-optimized tools directly into the IDE. That’s what makes workflows like: Linear → GitHub → PR → ticket updates actually executable end to end by an agent. Gateways expose tools built specifically for non-deterministic systems — consistent schemas, real OAuth, and predictable behavior — so agents don’t fall apart the moment they take action. Same models. Same prompts. Very different outcome. We put together a full write-up and demo showing this in action: https://lnkd.in/gJSJSV2Z

You're treating tools and skills like synonyms. Your architecture knows they're not. Tools = hands (execution, APIs, side effects) Skills = head (expertise, reasoning, judgment) Different costs. Different failure modes. Different security surfaces. Conflate them and you get: →Bloated context windows →Confused model selection →Auth nightmares in production The production lesson most teams learn too late: the protocol wars matter less than whether your agent can auth securely and execute reliably. We unpacked the real architectural tradeoffs 👇 Read the full breakdown: https://lnkd.in/gVFgttY2

Today the agent ecosystem levels up. The Linux Foundation has launched the Agentic AI Foundation (AAIF) with MCP, goose, and AGENTS.md as the founding open-source projects — and Arcade is honored to be a Founding Gold Member. Why it matters: - Open standards for how agents connect - Secure foundations for real-world actions - A unified community pushing the ecosystem forward Agents are moving beyond chat. They’re becoming infrastructure. And we’re excited to help shape what comes next. If you’re building multi-user AI agents that need secure, high-accuracy actions across any system, Arcade is the MCP runtime that makes it production-ready. Get started here: https://lnkd.in/geNa7mJ7

30-minute Arcade.dev breakdown just dropped on YouTube. What's covered: - Auth without building custom OAuth - 6,700+ production-ready tools - MCP servers that don't require heroic configuration - Real agent builds with CrewAI and Cursor No fluff. Just a straight walkthrough of how auth works when someone else handles the infrastructure. Props to Tyler Reed for the most thorough technical breakdown we've seen. Worth the time if you're building agents that need to actually do things. Watch the full breakdown → https://lnkd.in/gNhN3GjC

Anthropic just released Tool Search — a feature that lets Claude access thousands of tools without consuming its context window. We had to know: does it actually work at scale? So we loaded 4,027 tools and ran 25 straightforward tests. The kind of requests that should be automatic: - "Send an email to my colleague" - "Post a message to Slack" - "Schedule a meeting for tomorrow" - "Create a GitHub issue" We tested both search modes (regex and BM25) and measured whether the right tool even appeared in the results — before Claude tried to use it. The results surprised us. Some tool categories performed exactly as we'd hope. Others... didn't. This gives us real data about one of the hardest problems in agent infrastructure: how do you help AI reliably find the right tool when there are thousands to choose from? We published the full methodology, test results, and what this means for production agents: https://lnkd.in/gQy6eGa3 If you're building agents with large tool libraries, this is worth reading.

Today's MCP spec update just made agent auth actually work in production. We're not just celebrating this release — we helped write it. The biggest additions: → Secure OAuth flows via URL Mode Elicitation (we co-authored this) → OpenID Connect support → Incremental scope consent — finally, real least-privilege access → CIMD (Client ID Metadata Documents) — a simpler, modern way to register clients to MCP servers → Experimental long-running tasks for async server workloads → Better tool-naming rules, clearer error handling, and stronger validation → Formalized spec governance to steer MCP as it grows These aren’t minor tweaks. This is MCP growing up and preparing for production-grade agents. If you want the full technical breakdown — including how URL Elicitation actually works under the hood — William Dawson recorded a walkthrough that explains what changed and why it matters: https://lnkd.in/gjtqu_8e