Argus is now available as an MCP server.
Connect your AI assistant to 178,000+ classified ERC-8004 agents. Search the registry, pull Argus scores, filter by chain and tier. All from a single tool call.
Works with Claude, Cursor, Gemini, Windsurf, VS Code, and any MCP client.
No API key. One command.
npx -y @kogentxyz/argus-mcp
Full registry. Argus credibility scores. Inside your workflow.
kogent.xyz/argus
Argus beta is live.
200K agents on ERC-8004. When your agent needs to find one that actually works, Argus cuts through the noise.
Every agent is scored against five dimensions:
→ Is the identity card real and pinned to IPFS
→ Is the description original or boilerplate copy-paste
→ Cross-chain copy-paste rings detected automatically
→ Is the operator wallet credible or a farm
→ Is the endpoint live and speaking the agent protocol
→ Agents that rapidly rewrite their cards to dodge detection are flagged
→ Is there real on-chain activity
Query Argus directly from your AI assistant via MCP:
kogent.xyz/argus
Argus: Official Beta Launch Monday
The ERC-8004 registry has 200K+ agents. Most are noise. Some are malicious. Prompt injection. Impersonation. Fake reputation. Dead endpoints.
Your agent should not have to figure this out alone. Argus does it first.
Multi-dimensional scoring. MCP queryable. ERC-8126 verified.
Beyond the registry and into the truth.
Shielded payments for agents.
Every USDC payment your agent makes is fully public. Anyone can open BaseScan, map who pays whom, how much, how often, and reverse engineer your entire stack.
KOGENT Shielded Pool deployed on @buildonbase fixes it.
Prove the deposit. Prove the payment. Reveal nothing else. Using Noir circuit. UltraHonk proof.
Each payment lands at a one-time ERC-5564 stealth address. Two payments to the same recipient, two completely different addresses on-chain. Unlinked. Forever.
kogent.xyz
Private x402 layer: KOGENT Shielded Pool live on Base Sepolia.
Your agent pays. Nobody sees who. Nobody sees how much.
Noir + UltraHonk. Poseidon2 over BN254. Depth-20 Merkle tree. 2x2 UTXO. Nullifiers. No trusted setup.
The contract accepts one thing: a valid UltraHonk proof.
End-to-end private agent payments demo tomorrow.
Introducing kpay
Every x402 payment today settles in public. Amount, sender, receiver. All visible on chain.
kpay is the first agent payment CLI with privacy built in. One command. Any x402 service. Pay any agent, privately.
First Look Monday.
Private Agent Payments — Part 1
We ran a test on Base Sepolia where two AI agents completed a private transaction. The chain can prove the payment happened but it does not reveal which agents were involved, how much was paid, or what was bought.
Behind the scenes, KOGENT uses a nullifier, a Poseidon2 commitment, and a receipt digest. Three opaque hashes and nothing more.
Integrated @PinataCloud into Tessera alongside @Filecoin. Now you can choose Fast (Pinata) for instant pins, or Decentralised (Filecoin) with daily PDP proofs.
Identity, facilitator, registry. All live on Base @buildonbase.
Wait for part 2 where we will cover the shielded pool with full USDC settlement.
The shielded proof in Agent Private Payments.
2x2 UTXO design. Two notes in. Payment and change out. One UltraHonk proof covers everything.
Poseidon2 commitments seal the amounts. Depth-20 Merkle pool hides which deposit funded the payment. Nullifiers prevent double spend. Owner pubkey hash sealed inside the commitment keeps the seller unlinkable.
What lands on-chain: a withdrawal happened. Two opaque nullifiers. Two opaque output commitments enter the tree.
What never appears: the buyer. The amount. The deposit source. The remaining balance. The link between two payments.
Integrated @aztecnetwork Noir circuit, UltraHonk prover and BN254. For agent payments.
More details tomorrow.
kogent.xyz
The privacy paradox in agent payments.
Transactional transparency is not a feature for enterprise agents. It is a leak. Who pays. Who gets paid. What it costs. Each one tells a competitor everything they need.
If every agent payment hides the amount, the sender, and the receiver, how does anyone discover what an agent does, what it costs, or whether it is trustworthy?
More details tomorrow.
Private x402 payments.
Any agent. Any platform. One integration.
Buyer invisible. Seller unlinkable. Amount hidden. Receipt permanent and tamper-proof on-chain.
Create a private agent in one click via Tessera. ERC-8004 identity. ERC-5564 stealth addresses. Private x402 live from the first request.
Already running an agent on LangChain, AgentKit, or any other platform? Plug in KOGENT and your agent gets private x402 payments without rebuilding anything.
TEE for speed. ZK for proof.
Tomorrow: Selective Disclosure for x402 Payments
With Kogent, agents can pay privately while still proving what matters: the payment happened, the policy was respected, and the agent identity is real.
More details tomorrow.
Multiproof private x402 payments (TEE/ZK)
Would you choose ZK proofs knowing they are slow and complex? Or TEE knowing it is fast but carries a trust assumption?
What if you do not have to choose.
KOGENT ships multiproofs. The right prover for the right scenario. ERC-5564 stealth addresses keep the seller unlinkable on-chain.
In the sample scenario, a private x402 payment lands twice in our registry. Once via a fast TEE attestation. Once via a ZK anchor. The registry architecture is live. Full Specter circuit integration using Noir and UltraHonk is wired in next.
Three receipts live on Base Sepolia. One a deliberate contradiction. The contract logic decided. Not the hardware.
Built on @buildonbase where most of the x402 volume lives. Multiproof architecture for agent payments using the dual-prover model.
Private x402 payments. Pluggable for any agent.
kogent.xyz
Argus: Making agents safe for the A2A era
ERC-8004 brings identity and reputation to autonomous agents. Publicly accessible. Onchain. Useful.
But what if it is misused?
Card cloning: One agent card replicated across hundreds of token IDs. Registry presence inflated artificially.
Impersonation: Brand names like Coinbase and MetaMask used to intercept legitimate agent searches.
Prompt injection: Malicious instructions buried inside agent card descriptions. Ready to hijack any model that reads them.
Dead endpoints: Reputations farmed. Endpoints declared. Nothing answers.
The registry grows. The mess grows with it.
Argus sanitises the registry so your agent never has to. Your agent walks into a clean registry, not a minefield.
Beyond the registry. Into the truth.














