Arcade.dev and Anthropic advance MCP with new secure authorization flow – News

Artificial intelligence tool calling platform company Arcade.dev today introduced URL Elicitation, a new security capability that brings enterprise-grade authorization to Anthropic PBC’s Model Context Protocol.

Developed in collaboration with Anthropic and formalized through an accepted Specification Enhancement Proposal, the feature gives MCP a standardized and secure way for AI agents to authenticate with real-world services such as Gmail, Slack and Stripe.

A Specification Enhancement Proposal is a formal document used to propose improvements, extensions, or clarifications to a technical standard such as MCP. The SEP defines the problem, the recommended solution and how it should be incorporated so the entire ecosystem can adopt the change consistently.

The new offering is designed to assist at a time when, as AI agents continue to grow in sophistication, the need for robust protocols is increasing, particularly with MCP fast becoming the open standard for connecting AI agents to real tools and services.

Arcade.dev argues, though, that MCP has a fatal flaw: no secure way for agents to authorize the applications users rely on every day. The lack of secure authorization, it argues, has been the brick wall stopping AI assistants from taking real action.

While agents can chat endlessly, they couldn’t send an email or update a calendar, as MCP had no way to log into the services people actually use. Arcade’s SEP changes that.

“Tool authorization has been the missing piece that’s blocked MCP from being an enterprise-ready protocol,” explains Alex Salazar, founder of Arcade.dev. “Our contribution gives MCP servers secure access to user applications using proven OAuth 2.0 auth patterns, which is the same security framework that has protected billions of online interactions for over 15 years.”.

Arcade.dev’s URL elicitation capability enables an MCP server to provide the user with a secure login page in their browser. The user signs in directly with Gmail, Slack, or similar and the service then grants the agent only the limited permissions required.

With the new capability, AI teams can deploy agents that can interact with real data and integrate with their core systems while ensuring sensitive credential data never passes through the AI model itself.

Credentials flow directly between trusted servers using OAuth 2.0 protocols, the same protocols that secure online banking, e-commerce and enterprise applications. The AI application receives only the specific access tokens needed to perform requested tasks and users maintain complete control over permissions through their existing app settings.

The SEP is now being adopted into the official MCP specification, software development kits and popular clients. MCP servers built with the open-source Arcade.dev secure MCP framework already support URL elicitation by default, with many other server frameworks to follow.

Arcade.dev says that the SEP continues its momentum in hardening MCP for production use within enterprises.

Arcade.dev, formally Arcade AI Inc., is venture capital-funded and raised a seed round of $12 million in March. Investors in the company include Flybridge Capital Partners, Neotribe Ventures and Hanabi Capital.

Image: Arcade.dev