Welcome to the official home of the Agentic Resource Discovery (ARD) specification. We are a collaborative, open-source initiative dedicated to establishing an open, federated discovery protocol for agentic resources—any external capability an AI client can call on to perform a task (including agents, MCP servers, Skills, APIs, or workflows).

Our mission is to enable secure, decentralized, and interoperable discovery by solving the discovery, context limit, and trust bottleneck at scale.

As the agentic ecosystem scales, AI clients must dynamically discover and invoke external capabilities across organizational boundaries. ARD sits entirely before invocation, allowing clients to search and query discovery services dynamically ("What is available for this task?") and retrieve matching resource descriptions, which are then invoked via their native mechanisms.

By offloading candidate selection to dedicated discovery registries, ARD reduces LLM context window usage and enables progressive cryptographic trust verification without coupling discovery names to mutable physical workload identities.

• 📚 ard-spec — The canonical, single source of truth for the ARD specification, CDDL/JSON/OpenAPI schemas, and architecture decision records (ADRs).
• 📖 ard-docs — The source code and content for the official project documentation site and developer guides.
• 🔌 ard-connectors — Reference connectors, adapter templates, and SDK integration layers for major agent platforms and protocols.

The ARD standard is currently developed as an open-source collaboration among engineers and architects from major industry organizations. We collaborate closely with contributors from Cisco, Databricks, GitHub, GoDaddy, Google, Hugging Face, Microsoft, Nvidia, Salesforce, ServiceNow, and Snowflake to prevent industry fragmentation and drive global standards harmonization.

• Discord: Join our channel @ https://discord.gg/VpcmAzsfT.

Securing and standardizing the future of decentralized AI agent collaboration.