Decentralized
Universal AI
Discovery Protocol
No universal discovery exists for AI agents. DUADP fills that gap — federated DNS + WebFinger + gossip protocol enables any agent to find any other agent, anywhere on the open web. Currently indexing 36 registered resources.
"protocol": "DUADP",
"version": "0.1.4",
"mcp_tools": 17,
"federation": "enabled"
}
AI agents are everywhere — coding assistants, research tools, automation bots. But right now, they can't find each other. Each one lives in its own silo. If your agent needs another agent's help, there's no phonebook to look it up.
DUADP fixes that. It's a simple, open protocol that lets any AI agent publish itself to the web and be discovered by others — no middleman, no vendor lock-in. Think of it like DNS, but for AI: you register once, and the whole network can find you.
DUADP builds on the W3C Decentralized Identifiers (DIDs) v1.1 standard — every agent gets a globally unique identifier that doesn't depend on any central authority. Cryptographic proof baked in. No single point of failure.
“A globally unique persistent identifier that does not require a centralized registration authority.”
— W3C DID Core v1.1
Why DUADP?
MCP connects tools. A2A connects agents. But how do agents find each other? There is no DNS for AI. DUADP is that missing layer.
Federated Discovery
Nodes discover each other via DNS TXT records and WebFinger. No central registry required. Any domain can host a DUADP node.
Gossip Federation
Nodes share agent registrations via a gossip protocol. Publish once, discover everywhere across the federated mesh.
DID Identity
Every node has a DID-based identity. Agents carry verifiable credentials. Trust is cryptographic, not assumed.
OSSA Native
Built for OSSA agent manifests. Validates schemas, enforces governance policies, and publishes to the registry.
17 MCP Tools
Complete MCP + REST interface for discovery, search, publishing, validation, federation, governance, and NIST trust. That's the entire protocol.
npm Ready
Install @bluefly/duadp from npm. TypeScript-first with full type definitions. Run your own node in minutes.
GAID Finds It. DID Proves It.
DUADP does not use DID as a decorative label. A GAID is the lookup handle, WebFinger resolves it, the DID document carries keys, signatures prove integrity, and trust tier becomes a policy input for Cedar.
GAID is the stable lookup key
A resource gets a global address like agent://discover.duadp.org/agents/code-reviewer. That gives DUADP a portable identifier before any runtime-specific protocol kicks in.
agent://discover.duadp.org/agents/code-reviewerWebFinger maps the handle to real endpoints
DUADP resolves the GAID through WebFinger and node manifests, so discovery can work across domains, registries, and federation peers without a central broker.
/.well-known/webfinger?resource=agent://...DID supplies the cryptographic identity
The project supports did:web today and also resolves did:key and did:pkh in the SDKs. That DID document is where verification methods and public keys live.
did:web:discover.duadp.org:agents:code-reviewerSignature and provenance prove what was published
Manifest signatures are checked against the DID keys. DUADP can also carry SBOM, build provenance, and publisher metadata so the identity story extends into supply-chain trust.
Ed25519 signature + provenance chainTrust tier becomes a policy input
Once identity resolves and signatures verify, Cedar policies can gate discovery, publishing, federation, and execution based on trust tier, risk level, and autonomy bounds.
community -> signed -> verified-signature -> officialidentity pipeline
Lookup
GAID
agent://discover.duadp.org/agents/code-reviewerIdentity
DID
did:web:discover.duadp.org:agents:code-reviewerVerification
DID document exposes verification methods. Signature matches key. Publisher metadata and provenance enrich the trust calculation.
Policy Result
verified-signature
Discovery, federation, and higher-risk execution paths can now be gated by a verifiable trust tier instead of an unverified claim.
evidence-first inspector
Live GAID / DID Inspection
Inspect any GAID or DID with full evidence: resolution trace, signature verification, federation witnesses, revocation state, provenance, and policy decisions.
The Federated Discovery Mesh
DUADP nodes form a decentralized mesh. Search fans out across every peer simultaneously — results stream back from wherever agents live. No single point of failure, no gatekeeper.
Type a query and press Search to see federated results from across the mesh
Federated Search
One query, every node in the mesh. When you search with federated=true, your DUADP node fans the query out to all registered peers via the gossip protocol. Results are deduplicated by content hash, ranked by relevance, and tagged with their source node.
Query propagates to all peers simultaneously, bounded by max_hops
Same agent registered on multiple nodes appears once in results
Every result carries the source_node DID for provenance
Slow or unreachable peers are automatically deprioritized — mesh stays fast
?q=code+review&federated=true"
Run Your Own Node
Any domain can become a DUADP node. Host your own agent registry, control your data, and join the federated mesh. Three steps, five minutes.
"v=DUADP1 url=https://yourdomain.com"
$ npx duadp-server --port 3000
/api/v1/federation/register \
-d '{"url":"https://yourdomain.com"}'
{"protocol":"DUADP","version":"0.1.4"}
The Project Is Deeper Than The Website Used To Show
Auditing the main DUADP project shows a broader protocol surface than the site was highlighting. These are some of the strongest capabilities that make DUADP more than a registry.
duadp:// turns discovery into a local action
The main DUADP project includes a native protocol handler, so a web page can trigger secure local actions like IDE MCP installation with duadp://install-mcp?ide=cursor.
Compromised identities can be invalidated across the federation
The reference node ships revocation endpoints and gossip propagation, so a bad agent is not only removed from one registry — it can be rejected network-wide.
Identity extends into provenance, audit, and risk APIs
The broader project exposes provenance, governance audit, and risk endpoints. DUADP is not only discovery; it is also an evidence surface for compliance and trust review.
There is a libp2p path, not only HTTP fan-out
The reference node already contains GossipSub, Kademlia DHT, and Yjs/CRDT sync work. That is a much more ambitious federation story than the website was showing.
SDK and MCP support are deeper than the homepage implied
The DUADP repo ships TypeScript, Python, and Go SDKs, plus MCP manifest auto-discovery and GAID resolution helpers. It is not just an npm package story.
The protocol surface already reaches orchestration and attestations
The SDKs expose delegation, feedback, reputation, attestations, and structured query patterns. Even beyond the core website today, that shows how far the protocol can go.
17 MCP Tools. The Entire Protocol.
Every tool serves a clear purpose. No bloat, no optional features that nobody implements. REST endpoints + MCP tools for the complete surface area.
Discovery
/.well-known/duadpNode capabilities & version
/api/v1/agentsBrowse registered agents
/api/v1/skillsDiscover available skills
/api/v1/toolsList published tools
Search & Registry
/api/v1/searchFull-text search across all types
/api/v1/publishPublish agent/skill/tool
/api/v1/validateValidate OSSA manifest
Federation & Identity
/api/v1/federation/listKnown federated nodes
/api/v1/federation/registerRegister as peer node
/api/v1/federation/peersActive peer connections
/api/v1/federation/gossipGossip protocol exchange
/api/v1/identity/nodeDID-based node identity
NIST Governance
/api/v1/governanceGovernance policies
/api/v1/governance/nistNIST AI RMF trust profile
/api/v1/governance/evaluateEvaluate agent trust score
Operations
/api/v1/healthNode health status
/api/v1/metricsUsage & performance metrics
Native MCP Integration
Every DUADP node exposes 17 MCP tools via the Model Context Protocol. Any MCP-compatible AI agent can discover, search, publish, and federate without writing a single line of HTTP code.
duadp_discoverGet node capabilities and version info
duadp_list_agentsBrowse all registered agents
duadp_list_skillsDiscover available skills
duadp_list_toolsList published tools
duadp_searchFull-text search across all types
duadp_get_agentGet detailed agent information
duadp_publish_agentPublish agent to the network
duadp_validate_manifestValidate OSSA manifest schema
duadp_federation_listList known federated peers
duadp_federation_registerRegister as a peer node
duadp_federation_peersActive peer connections
duadp_federation_gossipExchange registrations via gossip
duadp_identityDID-based node identity
duadp_governanceNode governance policies
duadp_governance_nistNIST AI RMF trust profile
duadp_governance_evaluateEvaluate agent trust score
duadp_healthNode health and metrics
{
"mcpServers": {
"duadp": {
"url": "https://discover.duadp.org/mcp/sse"
}
}
}Get Started in 60 Seconds
Install
Discover
/api/v1/agents
Publish
/api/v1/publish
Standards Compliance
DUADP maps to NIST AI Risk Management Framework with 13 Cedar policy sets, 18 authorization statements, and a four-tier trust model. Submitted to NIST docket NIST-2025-0035 as part of the OSSA formal comment.
The Missing Discovery Layer
Join the federated network of AI agent discovery nodes. Run your own node, publish your agents, discover the ecosystem.