The authenticity mark for every AI decision.

A third-party certification layer for AI agents. Every decision is evaluated against your policy, scored for risk, signed, and anchored to Solana — so your customers, auditors, and regulators don't have to take your word for it.

newMCP-native — works with Claude, Codex, Copilot StudioMCP-native — drop into Copilot Studio, Claude, or any MCP agent EU AI ActArticle 12-ready audit logs a regulator can verify

Ed25519 signedSHA-256 hashedVerifiable offlineAnchored on Solana

Model Context ProtocolClaude CodeOpenAI CodexOpenClawAWS BedrockMicrosoft Copilot StudioAnthropic ClaudeLangChainCrewAIAutoGenOllamavLLMCustomModel Context ProtocolClaude CodeOpenAI CodexOpenClawAWS BedrockMicrosoft Copilot StudioAnthropic ClaudeLangChainCrewAIAutoGenOllamavLLMCustomModel Context ProtocolClaude CodeOpenAI CodexOpenClawAWS BedrockMicrosoft Copilot StudioAnthropic ClaudeLangChainCrewAIAutoGenOllamavLLMCustom

How a receipt is issued

Four steps. Every certificate.

A certification layer for AI agents — independent of the model or runtime. Every decision gets stamped, scored, and committed to chain.

Evaluate

Your policy runs against every decision — deterministic rules in code, plus an AI evaluator for fuzzy checks like injection, sensitive data, and unsupported claims.

Stamp

The verdict is captured into a structured receipt and signed with Ed25519. Tamper any byte and verification fails. This is the certificate itself.

Anchor

Every receipt's hash is committed to Solana automatically — independent, public timestamping. Off-chain payloads stay private; only fingerprints go on-chain.

Verify

Anyone, anywhere can re-hash the receipt, check the Ed25519 signature against your published key, and confirm the on-chain anchor. Trust doesn't require trusting us.

MCP-native

The native language of agents. Supported out of the box.

Agent Cert speaks Model Context Protocol — the open standard most modern agent runtimes already use to talk to tools. One endpoint, six tools, the same API-key auth as REST.

Plug into Microsoft Copilot Studio, Claude Desktop, or any MCP client in 30 seconds.
Every tool call is auth-gated by X-API-Key — same key system as REST.
Tools: certify_decision, get_receipt, verify_receipt, check_credits, list_my_receipts, get_anchor_proof.

claude-desktop-config.json

{
  "mcpServers": {
    "agent-cert": {
      "url": "https://api.agentcert.net/mcp/sse",
      "headers": {
        "X-API-Key": "ak_live_..."
      }
    }
  }
}

The certificate

A signed, on-chain trust record anyone can verify.

Every decision is stamped with what happened, which policy applied, what the evaluator concluded, and a verifiable signature. The issuer's public key is published; the hash is on Solana. Trust does not require trusting us.

Ed25519 signed · SHA-256 hashed
Anchored to Solana on every issue
Verifiable by anyone, anywhere, anytime
Compatible with any agent framework

decision-receipt.jsonsigned

1{
2  "receipt_id": "rcpt_01HXTRUST...",
3  "agent_id": "agent_invoice_001",
4  "platform": "bedrock",
5  "execution_hash": "sha256:abc123...",
6  "evaluation_hash": "sha256:def456...",
7  "trust_score": 91,
8  "risk_level": "low",
9  "policy_compliant": true,
10  "prompt_injection_detected": false,
11  "evaluator_version": "ai-evaluator-v1",
12  "policy_version": "policy_v1",
13  "created_at": "2026-05-18T22:09:08Z",
14  "signature": "ed25519:AUOCk1Ef3vH..."
15}

Decision Credits are funded by CERT.

Every verified receipt costs 3 credits. Credits are acquired only by sending CERT to the bridge — no fiat path, no exceptions. New accounts get 30 free credits to start.