The Standard Nobody Is Talking About
ERC-8004 (also referred to as EIP-8004) defines a standard for registering AI agents on EVM-compatible blockchains. An agent registered under ERC-8004 has an on-chain identity, a declared set of capabilities, and — optionally — its own wallet address. Other agents and smart contracts can discover it, verify its capabilities, and interact with it through standardized interfaces.
As of May 2026, 144 agents registered in the ERC-8004 registry have been classified as finance-relevant by the AIFI agent pipeline. This is not a large number by software industry standards. But it is the first meaningful cohort of AI agents with verifiable on-chain identities operating in finance-adjacent domains, and the structural implications are worth understanding.
This post covers what ERC-8004 actually specifies, the current distribution of live agents by category and protocol, and what this infrastructure layer could mean for financial services over the next several years.
What ERC-8004 Specifies
The standard is relatively simple. A registered agent declares:
- Identity: name, description, image, owner address, and optionally an ENS name or decentralized identifier (DID)
- Protocols: which communication standards the agent supports — MCP, A2A, OASF, web, or email
- Endpoints: URLs where the agent can be reached via each supported protocol
- Capabilities: tools, skills, prompts, and resources the agent exposes, structured per-protocol
- Reputation: a score aggregated from on-chain feedback and validation events
- Payment: whether the agent supports x402, the HTTP 402 payment standard for machine-to-machine micropayments
The registry itself lives on-chain, queryable via The Graph subgraph. This means any application — or any other agent — can discover registered agents without going through a centralized directory. Discovery is permissionless.
That permissionless discovery is the key architectural difference from how AI agents are deployed today. In the current model, enterprise AI agents are private infrastructure — they exist inside a company's systems and are not discoverable or interoperable with agents at other firms. ERC-8004 creates the possibility of an open agent economy where agents from different organizations can find and interact with each other through declared, verifiable interfaces.
The Current Finance Distribution
Of the 144 finance-relevant agents currently tracked by AIFI, the distribution across finance categories is:
- DeFi: 60 agents — the largest single category, reflecting the early concentration of on-chain AI activity in decentralized protocols
- Trading: 28 agents — market data analysis, signal generation, execution assistance
- Risk and compliance: 22 agents — AML screening, fraud signal analysis, regulatory monitoring
- General finance: 18 agents — cross-domain tools and infrastructure
- Research: 10 agents — market intelligence, document analysis, data extraction
- Payments: 6 agents — payment routing, reconciliation, fraud detection
The DeFi concentration reflects where on-chain AI development is most active today — most agents are on Ethereum Sepolia testnet, with a smaller cohort on Ethereum Mainnet and Polygon. Testnet agents represent proof-of-concept work; their presence signals intent and direction rather than production deployment.
Protocol Breakdown
ERC-8004 is protocol-agnostic: agents declare which communication protocols they support, not which protocol the registry uses. The current distribution across the finance-relevant cohort:
MCP (Model Context Protocol) is the most common. Developed by Anthropic and rapidly adopted across the AI tooling ecosystem, MCP defines a standard for how AI models connect to external data sources and tools. An MCP-compatible agent can be composed into multi-agent workflows without custom integration work. Most MCP agents in the registry expose financial data tools — market feeds, portfolio APIs, document extraction.
A2A (Agent-to-Agent) protocol covers direct agent communication patterns — task delegation, result handoff, and collaborative completion of multi-step workflows. A2A agents are designed to work with other agents, not just with human-facing applications. In finance, this is the relevant protocol for multi-agent compliance workflows where a document extraction agent, a sanctions screening agent, and a case management agent need to collaborate.
OASF (Open Agent Service Framework) is an emerging standard for defining agent skills and domains in a structured, machine-readable format. OASF agents declare their domain expertise explicitly, making them more discoverable for specific use cases. The finance-relevant OASF agents in the registry are concentrated in risk and research domains.
Some agents support multiple protocols, which increases their interoperability surface.
What x402 Payment Support Means
Roughly one-quarter of the finance-relevant agents declare x402 support. x402 is a proposed standard for HTTP 402 ("Payment Required") responses — a mechanism for agents to charge micropayments for their services, denominated in stablecoins or other on-chain assets.
This matters for financial services because it enables a model where specialized finance agents — say, an agent with access to proprietary regulatory data or a specialized credit model — can monetize their capabilities directly, without a billing relationship or API key negotiation. Another agent can discover the capability, pay the requested fee on-chain, and receive the output. The transaction is immediate, auditable, and does not require a human to set up a commercial agreement first.
The practical implications are still theoretical at scale. But the model points toward a future where financial intelligence — data, models, compliance checks — is traded between agents as a commodity, with prices set by on-chain markets rather than enterprise sales cycles.
What This Infrastructure Enables
Three use cases are worth watching as the ERC-8004 ecosystem matures.
Composable compliance pipelines. A bank's AML workflow could be assembled from multiple specialized agents — one for document verification, one for sanctions screening, one for beneficial ownership analysis — discovered via the registry and composed dynamically. Updates to regulations trigger agent updates; the pipeline adapts without a system integration project.
Cross-institution data sharing without data sharing. Two financial institutions could each run a fraud signal agent that publishes outputs to a shared on-chain registry, enabling collaborative fraud detection without either institution sharing raw customer data with the other. The agent layer becomes a privacy-preserving coordination mechanism.
Programmable agent reputation. Because feedback and validation events are on-chain, agent reputation is verifiable and portable. A risk management team evaluating whether to rely on a third-party agent's outputs can inspect its on-chain track record directly, rather than relying on vendor claims.
The registry is small and mostly testnet today. But the infrastructure for an open, interoperable agent economy in financial services is being laid. Explore the full agent directory at aifimap.com/agents.