Agent CertAgent Cert
Documentation

Agent Cert docs.

Everything you need to certify, anchor, and verify AI agent decisions. Pick your path below — or jump straight to the quickstart.

Pick your path

Quickstart

Get your first signed, anchored receipt in 5 minutes. Sign up, copy your API key, curl one endpoint.

Read
REST API

Full reference for /v1/receipts, /v1/credits, /v1/auth. Authentication, request/response schemas, status codes.

Read
MCP Server

Plug Agent Cert into any MCP-capable agent (Copilot Studio, Claude Desktop, custom). Six tools, one endpoint.

Read
Integrations

Platform-specific guides: Microsoft Copilot Studio, AWS Bedrock, LangChain, CrewAI, custom agents.

Read
Verification

How public verification works. Signatures, hash chain, Solana anchor proof. Anyone can verify any receipt.

Read
CERT & credits

How Decision Credits work, how to fund them with CERT, the locked 1 : 100 bridge rate, the Free Trial.

Read

What is Agent Cert?

Agent Cert is a third-party certification network for AI agent decisions. Every meaningful action your agent takes can be submitted to our API, where it's evaluated against your policy, scored for risk, signed with Ed25519, and anchored on Solana. The output is a Decision Receipt — a portable, tamper-evident record anyone can verify, with no account required.

Three guarantees

  • Signed. Every receipt carries an Ed25519 signature from Agent Cert's public key.
  • Anchored. Every receipt's hash is committed to Solana. Public ledger, public memo.
  • Verifiable. Anyone — your customer, your auditor, your regulator — can verify a receipt at agentcert.net/r without an account.

How it fits together

Every receipt follows the same seven-step flow. Steps 1–6 happen server-side on a single API call (~500–2000ms). Step 7 is public verification anyone can run, anytime.

Server-side · one API call

  1. Your agent runs

    01

    Any runtime: Claude Code, OpenAI Codex, Microsoft Copilot Studio, AWS Bedrock, LangChain, CrewAI, OpenClaw, or a custom agent.

  2. Submit execution to API

    02

    POST /v1/receipts — or call certify_decision over MCP. Same auth (X-API-Key), same data model.

  3. Policy check (deterministic)

    03

    Your bound rules run first — cheap, fast, in-code. Required-field checks, allowlists, dollar caps, hash validation.

  4. AI evaluation

    04

    Independent scoring for policy intent, prompt-injection risk, sensitive-data exposure, and unsupported claims.

  5. Ed25519 signature

    05

    The receipt is sealed by Agent Cert's signing key. Any tampering after this point breaks verification.

  6. Solana hash anchor

    06

    Receipt + evaluation hashes committed to Solana via a durable background worker. Public ledger, public memo, no private data on-chain.

  7. Decision Receipt issued

    07

    Returned to your agent in the same call — receipt_id, trust_score, risk_level, signature, anchor URL, verify URL.

Public · anyone, anytime

Verifiable forever at /r

07

Anyone — your customer, your auditor, your regulator — can verify any receipt at agentcert.net/r with no account required.

What it costs

Every receipt costs 3 Decision Credits. New accounts get 30 free credits at signup — enough to issue 10 certified receipts. After that, credits are acquired by sending CERT to the treasury wallet at a locked 1 CERT = 100 credits rate. There is no fiat payment path.

Full economics: certnetwork.net