DNS-AID: Linux Foundation Launches Decentralized Discovery for AI Agents

The Linux Foundation has launched DNS-AID, an open-source protocol that leverages existing DNS infrastructure to enable decentralized discovery and verificatio…

On May 27, 2026, the Linux Foundation announced DNS-AID, an open-source project that transforms the existing Domain Name System (DNS) into a discovery infrastructure for artificial intelligence agents and Model Context Protocol (MCP) servers. Moving away from current models—which often rely on hardcoded URLs or vendor-managed registries—DNS-AID establishes a decentralized, implementation-agnostic, and web-native hierarchy. Initially developed by Infoblox, the project debuts with three core reference components: a Python SDK, a CLI, and an MCP server.

Key Takeaways

DNS-AID utilizes the existing DNS framework as a global directory for AI agents and MCP servers, removing dependencies on centralized registries and fragile configurations.
The protocol is implementation-agnostic, functioning with any DNS provider regardless of vendor or technology stack.
The initial release includes a Python SDK, a command-line interface (CLI), and an MCP server, available immediately for implementation.
Originally developed by Infoblox, the project has transitioned to open governance under the Linux Foundation, led by Jim Zemlin and Ingmar Van Glabbeek.

From Hardcoded URLs to DNS Hierarchy: The Problem DNS-AID Solves

As AI agents proliferate, they are increasingly tasked with orchestrating workflows, querying APIs, and negotiating complex tasks autonomously. However, their ability to interoperate depends on their ability to locate one another. Current agent connectivity solutions are fragmented, often relying on brittle configurations such as hardcoded endpoints or directories maintained by individual operators. This architecture creates silos, vendor lock-in, and single points of failure whenever a registry changes its policies or goes offline.

DNS-AID addresses this by shifting discovery to a different layer. By leveraging globally distributed DNS infrastructure, the protocol allows every agent to publish its identity and capabilities within an existing hierarchy, providing the same reliability that DNS brings to standard domain name resolution. The system is strictly implementation-agnostic; it does not mandate a specific DNS provider or a proprietary record format.

Infoblox Contribution and Linux Foundation Governance

The project originated at Infoblox, a specialist in DNS, DHCP, and IP address management. The decision to donate the code to the Linux Foundation marks its transition from a corporate product to common infrastructure. As of May 27, 2026, the project is under open governance, and the community is invited to contribute to the development of the protocol and its reference implementations.

The released reference implementation consists of three components: a Python SDK for integration into existing applications, a CLI for manual operations and debugging, and an MCP (Model Context Protocol) server that exposes discovery functionalities.

"AI agents are quickly becoming the connective tissue of the modern internet, but without secure, open discovery infrastructure, that connectivity becomes a liability" — Jim Zemlin, CEO Linux Foundation

The "Web-Native" Model and Maintainer Insights

Ingmar Van Glabbeek, project maintainer for DNS-AID, describes the project’s objective as a shift toward a "web-native" model for artificial intelligence. According to Van Glabbeek, leveraging the existing DNS hierarchy allows developers to publish and discover agents with the same reliability and ubiquity that has powered web navigation for decades.

Choosing DNS as an identity layer provides an immediate architectural advantage: the system is already decentralized, scaled for billions of daily queries, and managed by a diverse range of institutional and commercial actors. While the project aims to allow agents to "verify each other," the source does not yet detail the specific cryptographic verification mechanisms, DNS record types, or extensions utilized to achieve this.

Operational Impact

For developers and AI platform operators, DNS-AID introduces three immediate changes. First, it enables the publication of agents without reliance on a centralized registry, mitigating the risk of service disruption if a vendor alters terms or ceases operations. Second, the availability of the Python SDK and CLI allows for immediate integration into test environments, with the MCP server ready to expose discovery features to compatible frameworks. Third, the transition to Linux Foundation governance opens the project to external contributions, removing it from the exclusive control of Infoblox.

For organizations managing DNS infrastructure, this shift introduces a new category of records and queries to monitor. As AI agents begin to populate the DNS with identity and capability data, operations teams must evaluate how this data interacts with existing zone management policies, TTL (Time to Live) settings, and propagation requirements.

Frequently Asked Questions

Is DNS-AID an IETF standard or a community project?

It is an open-source project under the governance of the Linux Foundation. It is not currently a standard ratified by the IETF or any other formal regulatory body, and there is no indication of a formal standardization process currently underway.

How does DNS-AID differ from a central AI agent registry?

DNS-AID eliminates the centralized model. Each agent publishes itself within the existing DNS hierarchy, utilizing a distributed infrastructure controlled by multiple actors rather than a single database managed by a specific vendor or consortium.

Does the protocol require changes to existing DNS infrastructure?

No. The protocol is implementation-agnostic and works with any DNS provider, regardless of the technology stack or vendor used.

What components are available to start using DNS-AID?

The reference implementation includes three components: a Python SDK, a CLI, and an MCP server. These are distributed under an open-source license for immediate experimentation.

Sources

Information has been verified against cited sources and is current at the time of publication.