Summary
Arcade.dev, a startup specializing in integrating AI agents with business applications, has launched a new Tool Recommendation system designed to cut firms' input token costs by up to 37%. This innovative feature allows AI agents to find the appropriate actions to take without exhaustively loading extensive lists of software functions, thereby significantly reducing overhead during interactions. Arcade’s system not only improves efficiency in managing AI tool usage but also operates alongside existing runtime controls for permissions and policy enforcement, setting it apart in a competitive landscape that includes Microsoft, Anthropic, and OpenAI, which also offer tool discovery mechanisms. The Tool Recommendation feature, which becomes available on October 15, aims to streamline workflows while ensuring security and compliance standards are maintained, further enhancing Arcade's offerings for enterprise customers.
Analysis
MCP: MCP, or Model Context Protocol, is an emerging standard for securely linking AI agents and models to external applications and tools. Arcade contributes to its development by authoring tool authorization specifications and participating in steering committees on MCP security and governance. The protocol supports user-specific permissions through OAuth flows, enabling agents to act only within authenticated scopes across enterprise systems. Arcade: Arcade.dev is a startup that provides an actions runtime for enterprise AI agents, focusing on secure connections to business systems through the Model Context Protocol (MCP). The company supplies infrastructure for authentication, authorization, policy enforcement, and tool execution while maintaining visibility and control for administrators. Its recent Tool Recommendation feature builds on this runtime to streamline tool discovery for agents operating in environments like Google Workspace, Slack, and Salesforce. Sam Partee: Sam Partee is co-founder and chief technology officer at Arcade, where he has led development of the company's security, oversight, and recommendation technologies. He presented Arcade's approach to agent permissions and visibility at the VentureBeat Transform 2026 Innovation Showcase. Partee has emphasized building on data from agent actions to train models that predict relevant tools across common enterprise tasks. Arcade Cloud: Arcade Cloud is Arcade's fully managed SaaS platform that hosts the MCP runtime and related capabilities for production AI agent deployments. It includes pre-built tool catalogs, security controls, audit logging, and the new Tool Recommendation system, which became available to customers on October 15 without an additional feature charge. The platform supports integration with frameworks like LangChain and models such as those from OpenAI and Anthropic while allowing custom policies and hooks. Tool Recommendation: Tool Recommendation is a feature from Arcade that allows AI agents to discover and rank relevant tools on demand instead of loading extensive catalogs into their context windows. It leverages smaller specialized models to match agent needs with actions from a catalog exceeding thousands of options, then executes selected functions through the MCP gateway. The system integrates with existing agent frameworks and LLMs to reduce overhead in multi-system workflows without altering core agent intelligence. Enterprise Focus: Arcade combines tool recommendation with runtime controls for permissions, policy enforcement, and auditing, distinguishing its offering in the market for secure agent deployments. Competitive Landscape: Microsoft, Anthropic, and OpenAI provide tool discovery mechanisms for their AI agents and models that defer loading full definitions until needed. Technology Flexibility: The Tool Recommendation system operates across multiple LLMs, agent frameworks, and MCP-compatible applications without requiring customers to rebuild integrations.