source: arxiv artificial intelligence: ai tool discovery at scale: all you need is dns
level: research
ai agents need to find and use many tools, but current discovery methods are slow and rely on central authorities. a new system called tooldns adapts the domain name system to handle tool discovery. dns is already a global, decentralized infrastructure that resolves names quickly. by mapping tool functions to dns names, tooldns turns a complex semantic search into simple name lookups. this reduces the time to find a tool from linear to logarithmic complexity.
tooldns adds three features to standard dns to support tool discovery. partially unfolded names allow flexible matching of tool capabilities. edns0 intent payloads carry information about what the agent wants to do. logical subdomains organize tools by trust and function. these changes let the system prune irrelevant tools early, so agents only see options that match their needs and permissions. the approach works within existing dns protocols, avoiding the need for new infrastructure.
the researchers built a benchmark with 33,688 real-world tools to test tooldns. this dataset covers many types of tools and helps measure how well the system scales. by using dns, tooldns inherits its resilience and distributed nature. it avoids the bottlenecks of centralized registries and can handle millions of tools. the method shows promise for enabling autonomous agents to discover tools at internet scale without sacrificing speed or reliability.
why it matters: it offers a scalable, decentralized way for ai agents to find tools, which is critical as the number of available tools grows.
source: arxiv artificial intelligence: ai tool discovery at scale: all you need is dns