Arcade.dev | The MCP Runtime

The gap comes down to design. Composio proxies the raw API → full records + metadata. Arcade returns only what the agent asks for At scale, this translates to ~$2M/year in token costs. Full benchmark is open source. arcade.dev/blog/attio-mcp…

Tool quality has a price tag. We benchmarked our Attio MCP toolkit against Composio's across 8 identical CRM queries. Same workspace, same model. Both got the right answers. Arcade: 7,426 tokens Composio: 747,083 tokens ~100x difference.

Better tools mean better agents. We're releasing this openly because the whole ecosystem wins when the bar goes up. Go see how your tools score 👇 toolbench.arcade.dev

The criteria are grounded in what matters. Definition quality draws on our 54 Agentic Tool Patterns (arcade.dev/patterns). Protocol compliance draws on Nate Barbettini's MCP Debugger. Security on OAuth 2.1 and related RFCs. Supportability on signals of project health.

We analyzed 218,426 MCP tools across 41,902 servers. 99.5% failed to earn an A or above. 76.6% got an F. The tools are where agents go to die. Here's what we found 🧵

Instead of calling tools one at a time, TanStack AI's Code Mode has the LLM write a TypeScript program that orchestrates all the calls in a sandbox. Same query, 4K context instead of 17K. @nearestnabors talked to @jherr about it: youtu.be/DKkE_-Ky3AA

Coding agents already know how to use the CLI. So why are we building elaborate MCP integrations instead? Our Head of Engineering built mcpx to answer that. curl for MCP. arcade.dev/blog/curl-for-…

@daytonaio More: arcade.dev/blog/daytona-s…

Wrapped up a great few days with the @daytonaio team in San Francisco. The Daytona toolkit by Arcade lets agents clone repos, run tests, open PRs, and execute full development workflows inside isolated Daytona sandboxes, with full autonomy and zero risk to production.

Every web developer's agent workflow: copy console error, paste into agent, wait, repeat. Chrome DevTools MCP fixes this: ~25 tools that give your coding agent direct access to the browser. @nearestnabors talked to @MHablich about how they built it. youtu.be/CIOcjjgl9MI

A common use case for agents is to ingest content, use the LLM to prioritize it and compile a digest of only things relevant to you. I'll walk you through how to build this agent so you're never trapped into clickbaits again! See you at 13:30 PST: youtube.com/live/oUY3VWCzP…

Unsolicited text from a dev who just set up Arcade for his OpenClaw instance: "10,000,000% easier than Auth0 token management. You guys will go far. Weird to know that logically and then experience it." Sometimes you have to feel the difference.

Read more about the toolkit: daytona.io/dotfiles/sandb…

Showing the @daytonaio toolkit by Arcade live at Daytona Compute today. 46 agent-optimized tools covering the full sandbox lifecycle: code execution, file ops, Git, SSH with GitHub OAuth built in so agents never have to manage a PAT. Come find us if you're at the show.

Full agenda: #agenda" target="_blank" rel="nofollow noopener">compute.daytona.io/#agenda

Catch us live at Daytona Compute this Sunday! Our co-founder & CEO Alex Salazar will be on the main stage demoing the Daytona sandboxes, now available as Arcade tools. We’re hyped about this partnership and can’t wait to show what we’ve been building together 🤝

Going live in 15 minutes to explain contextual access for tool execution! Join me at @TryArcade's YouTube channel! youtube.com/live/8Rb_KmQhB…

We wrapped this edition of MCP Connect Day with 3 fantastic talks from @OpenAI, @github and @TryArcade Thanks to all our speakers for sharing their MCP stories and insights!