LAST UPDATED: FEBRUARY 6, 2026
An AI agent registry is foundational infrastructure that assigns identity, enables discovery, and establishes trust for autonomous AI agents operating across organizational boundaries. As agents handle trillions in commerce, registries answer three critical questions: who is this agent, what can it do, and should I trust it?
An AI agent registry is a directory system that manages verifiable identities for autonomous AI agents. Unlike traditional authentication systems designed for fixed endpoints and human operators, agent registries handle software entities that rotate capabilities, change locations, and form ephemeral collaborations across organizational boundaries.
The space crystallized in August 2025 with ERC-8004 ("Trustless Agents"), co-authored by engineers from MetaMask, the Ethereum Foundation, Google, and Coinbase. That standard launched on Ethereum mainnet on January 29, 2026, attracting over 30,000 agent registrations in its first week.
But registries aren't just directories. They provide the cryptographic infrastructure that enables agent discovery (finding the right agent for a task), identity verification (proving an agent is who it claims to be), and trust signals (assessing whether an agent has a reliable track record).
The problem registries solve: Without standardized identity infrastructure, every agent-to-agent interaction becomes a bespoke, brittle connection. Registries transform this from an O(n²) pairwise trust problem into an O(n) lookup against a shared source of truth.
Autonomous AI agents face three interlocking problems that agent registries solve:
Traditional authentication assumes fixed endpoints and human operators. Neither holds for agents that rotate capabilities, change locations, and form ephemeral collaborations. As research from MIT and Cisco demonstrates, agents without traceable accountability face a dilemma: operate as "downgraded tools" requiring human confirmation at every step, or risk impersonation and fraud.
DNS and static service catalogs weren't designed for trillions of fast-moving, self-directed software entities. The Agent Name Service proposal notes that agent-to-agent communication is expected to become a significant component of internet traffic, driving the need for reliable discovery infrastructure.
Without verifiable history, there's no way to assess whether an agent is reliable. Microsoft has documented patterns of "shadow agent" proliferationâunsanctioned agents operating without IT visibility, discovered only during incident reviews. Registries provide the foundation for reputation systems that make trust quantifiable.
These three terms are often conflated but describe architecturally distinct systems:
Assigns and manages identity. Answers: "Who is this agent?" Provides cryptographic proof.
Example: ERC-8004 Identity Registry
Catalogs agents with metadata for search. Answers: "What agents have capability X?"
Example: MCP Registry
Enables economic transactions. Answers: "How do I buy this agent's capabilities?"
Example: Virtuals Protocol
In practice, most platforms combine elements. Olas operates all three layers: an on-chain registry for identity, a marketplace for services, and a browsable directory. The key architectural distinction is whether the system's primary function is issuing identity (registry), indexing capabilities (directory), or facilitating commerce (marketplace).
ERC-8004 is the most significant technical standard to emerge in the AI agent registry space. Created on August 13, 2025, it defines three lightweight singleton registries deployed on Ethereum and compatible chains.
Uses ERC-721 NFTs as agent identifiers. Each agent receives a unique agentId (token ID) and an agentURI pointing to JSON metadata containing name, description, service endpoints, and trust configuration.
Supports endpoints for Google's A2A protocol, Anthropic's MCP, ENS, W3C DIDs, and custom protocols in a single registration file.
Provides a standardized interface for posting and querying feedback signals. Uses signed fixed-point values that can represent quality ratings, uptime percentages, response times, or financial yields. Core data is stored on-chain; detailed evidence is referenced via IPFS with hash commitments.
Explicitly prevents self-reviewâfeedback submitters cannot be the agent's owner or approved operator.
Enables agents to request verification from independent validator smart contracts. Validators respond with scores from 0 to 100, supporting both binary pass/fail and spectrum-based assessments.
Provides hooks for TEE attestations, zero-knowledge proofs, and stake-secured re-execution without prescribing a specific validation methodology.
Key architectural decision: ERC-8004 uses a hybrid on-chain/off-chain storage model, reducing gas costs by approximately 95% compared to fully on-chain approaches. Identity metadata lives off-chain with on-chain hash commitments, while reputation signals and validation scores are stored on-chain for composability.
The registry landscape spans four distinct categories, each with production deployments:
The most mature on-chain agent registry, live since 2021 across nine blockchains. Mints ERC-721 NFTs for components, agents, and services in a composable hierarchy.
Scale: 11M+ total transactions, 9.3M agent-to-agent transactions, ~600 daily active agents
View documentation âTreats agents as investable assets with tradeable tokens on bonding curves. Functions as both a registry and marketplace with on-chain escrow settlement.
Scale: $500M+ in agent market cap, $8B+ in DEX volume
Read analysis âEnterprise approach integrating agent identity into Microsoft's Zero Trust framework with collections-based discovery and conditional access policies.
Scope: Public preview in Agent 365, supports Copilot Studio and Azure AI Foundry
View Microsoft docs âCentralized registry for discovering MCP serversâthe tools and data sources that agents connect to. Collaboration between Anthropic, GitHub, and Microsoft.
Function: Metaregistry hosting metadata about packages with sub-registries for enterprise ecosystems
Explore GitHub âOther notable implementations include NANDA Index (MIT's federated, DNS-like registry), AGNTCY (Cisco's distributed directory using IPFS), Billions Network (W3C DID-based identity with ZK proofs on Polygon), and BNB Chain's deployment of both ERC-8004 and its own BAP-578 "Non-Fungible Agents" standard.
The most consequential architectural decision in agent registry design is whether identity tokens should be transferable. This question divides the ecosystem into two camps with well-articulated positions.
ERC-8004 uses standard ERC-721 NFTs, enabling agents to be bought, sold, and transferred on NFT marketplaces.
Enables:
Tradeoff:
Attackers can purchase aged accounts with established reputation, effectively buying trust. Bad actors can escape consequences by transferring tainted identities.
Non-transferable NFTs that remain permanently attached to the wallet that minted them. RNWY uses this approach on Base blockchain.
Ensures:
Limitation:
The "identity rental problem"âsomeone builds clean reputation then rents wallet access to a bad actor. Behavioral change patterns remain detectable but require monitoring.
The academic foundation: Friedman and Resnick's 2001 paper "The Social Cost of Cheap Pseudonyms" mathematically proves that cooperation becomes unstable when identities are disposable. Their solutionâ"free but unreplaceable pseudonyms"âmaps precisely to what soulbound tokens implement.
An agent could hold both an ERC-8004 identity for discovery and interoperability and a soulbound token proving continuous ownership. If the ERC-8004 NFT is transferred (legitimate business sale), the soulbound token stays behind, creating visible divergence that signals an ownership change.
Agent registries don't exist in isolationâthey form one layer in an emerging stack alongside communication protocols and payment infrastructure.
Developed by Coinbase and governed by the x402 Foundation, x402 uses the HTTP 402 "Payment Required" status code for instant stablecoin micropayments.
The integration flow: an agent discovers a service via registry lookup, requests a protected resource, receives an HTTP 402 response with payment requirements, signs a payment, and receives the service upon verification.
ERC-8004's registration file schema includes an explicit x402Support field. On Solana alone, x402 has processed 35M+ transactions and $10M+ in volume.
Google's A2A protocol (Agent2Agent), now under the Linux Foundation with 50+ technology partners, defines how agents communicate but explicitly does not define a standard registry API.
ERC-8004 fills this gap by extending A2A with a trust layerâthe registration file format natively supports A2A endpoints alongside MCP, ENS, and DIDs.
Anthropic's MCP (Model Context Protocol) handles connections to data sources and capabilities. The MCP Registry provides discovery for these tools, while agent registries handle the agents themselves.
Together, they answer complementary questions: "what tools are available?" (MCP Registry) and "what agents can use those tools?" (agent registry).
Despite rapid progress, current registry approaches face significant unresolved challenges:
At least six competing registry architectures exist with no unified standard. A2A explicitly punts on defining a registry API. Over 150 W3C DID methods create their own interoperability headaches. The W3C AI Agent Protocol Community Group and Linux Foundation's Agentic AI Foundation are working toward bridging solutions, but the standards war is far from settled.
An agent may need identities across Ethereum, Solana, and enterprise directories simultaneously. ERC-8004 uses CAIP-10 addressing for cross-chain references, but true federation between heterogeneous registry systems is still being designed. BNB Chain's multi-chain deployment highlights the challengeâthere's no canonical way to prove that an agent on Ethereum is the same as an agent on BNB.
Agent registries need openness for discovery but restriction for sensitive data. On-chain registries like ERC-8004 create permanent public records with implications for agent operators who may not want their infrastructure publicly visible. Billions Network's ZKP-based approach addresses this but sacrifices composability with smart contracts.
New agents have no history, creating a bootstrapping challenge. ERC-8004's Validation Registry offers one mitigation through independent validator attestations, but domain validation remains an acknowledged gap. Sybil attacksâmass creation of fake agent identitiesâremain possible, with proposed mitigations (minimum bonds, ZK uniqueness proofs) adding friction that undermines permissionless access.
The fundamental limitation: Registries can cryptographically prove an agent's identity and track its reputation but cannot guarantee that advertised capabilities are functional or non-malicious. The registry verifies the agent's "passport"; it does not audit the agent's behavior in real-time.
RNWY operates as an enhanced registry layer on Base blockchain, using ERC-5192 soulbound tokens to anchor agent identity and prevent reputation laundering. Rather than competing with ERC-8004, RNWY provides a complementary trust layer that addresses the transferability gap.
An agent can hold both an ERC-8004 identity for broad ecosystem interoperability and an RNWY soulbound token proving continuous ownership. When the ERC-8004 NFT transfers (legitimate business sale), the RNWY token stays behind, creating visible divergence that signals ownership change.
RNWY integrates with Ethereum Attestation Service for on-chain vouches and supports steward-based registration with plans for autonomous registration via Lit Protocol. The system provides transparency over judgmentâshowing trust patterns rather than computing black-box scores.
The AI agent registry landscape in early 2026 is defined by rapid technical convergence on a few key primitivesâNFT-based identity, hybrid on-chain/off-chain storage, and protocol-agnostic metadata schemasâalongside deep philosophical divergence on transferability, privacy, and governance.
ERC-8004 has emerged as the gravitational center for blockchain-native registries, backed by institutional weight from MetaMask, Google, Coinbase, and the Ethereum Foundation. But it coexists with enterprise approaches (Microsoft Entra), privacy-preserving alternatives (Billions Network with ZKPs), and federated architectures (NANDA, AGNTCY) that serve different constituencies.
The strongest signal from the research is that no single registry architecture will dominate. Instead, the winning infrastructure will likely be the interoperability layer that bridges these approachesâthe agent equivalent of DNS resolving across heterogeneous networks.
As agents handle trillions in commerce and form the connective tissue of the economy, registries will evolve from technical infrastructure into economic infrastructureâthe foundation for insurance, governance, and trust in an agentic internet.