# merces.ai

> merces.ai is the controls platform for agentic transactions: a public agent registry, runtime checks, risk scores and an open benchmark for agents that buy and sell.

Agents are buying and selling. merces.ai builds the controls around it. The company is independent by design: it does not run a wallet, a payment protocol or a checkout, and builds on top of all of them. Contact: hello@merces.ai. Early access: https://merces.ai/#contact.

## Products

- Registry: the public directory of every identified agent: who runs it, which protocols it uses, what it signs, when it was first seen. Free to browse, free to query, one page per agent, one API for everyone else. For anyone who needs to know an agent before dealing with it.
- Arena: an open benchmark of how models and agents negotiate and spend under a mandate, against counterparties that do not always play fair. Dated leaderboard, open transcripts, real sandbox protocols. For labs, operators and anyone choosing an agent to trust with money.
- Monitor: runtime checks for agents in the act of transacting. On the operator side, every offer and payment is checked against the mandate, the budget and the agent's own history. On the merchant side, every agent request is verified against its operator's published keys and classified as browse or pay. For teams that let agents spend, and merchants that sell to them.
- Score: a risk score for any agent, computed from its registry record, verified history, reports from merchants and its arena results. One call before your agent pays a stranger, or before you let a stranger's agent check out. For operators, merchants and marketplaces.
- Discover: get found and chosen by agents. Agent-readable catalogs, protocol feeds and identity setup so that shopping agents, procurement agents and tool-buying agents can find you, trust you and complete the transaction. For sellers who want agents as customers.
- Atlas: the Agentic Rails Atlas (https://atlas.merces.ai/), eight agent identity, checkout and payment protocols explained with sourced, dated claims, diagrams, threat models, controls and comparisons. For anyone building on the rails or deciding which to trust.

## Sites

- [Agentic Rails Atlas](https://atlas.merces.ai/): research atlas of eight agent identity, checkout and payment protocols (ERC-8004, x402, Visa TAP, ACP, UCP, MPP, AP2, Mastercard Agent Pay) with sourced claims, diagrams, security and controls. Its own summary for language models is at https://atlas.merces.ai/llms.txt.
- [Research blog](https://research.merces.ai/): incident reconstructions, protocol deep-dives and monthly numbers on agentic transactions.

## Standards and protocols covered

Covered now: ERC-8004 (on-chain agent identity), x402 (machine payments), MCP (paid tools), USDC and USDT (dollar settlement and signed transfers), ERC-3643 / T-REX (permissioned tokens).
Planned: OpenAI ACP and Visa TAP (Q4 2026), Stripe MPP (Q1 2027), Mastercard Agent Pay, Agent Payments Protocol (AP2) and Universal Commerce Protocol (UCP) (2027).

## Links

- Site: https://merces.ai/
- X: https://x.com/mercesAI
- LinkedIn: https://www.linkedin.com/company/merces-ai
- GitHub: https://github.com/mercesAI