Loading...
Loading...
Agent Integrations
Drop OmegaEngine into any AI agent — in one line. Add the Model Context Protocol server to Claude Desktop, Cursor, Windsurf, and the rest, and your agent can govern, red-team, and verify the signature on every action it takes — before it executes.
Proof, not promises. Every verdict is signed and tamper-evident. The verify_prooftool checks a decision's integrity with OmegaEngine's server; for offline, trustless verification, check the Ed25519-signed attestation with @omegaengine/verify — no trust in OmegaEngine required.
The hosted remote endpoint serves all 8. The stdio package npx -y @omegaengine/mcp ships the 4 core action tools — remote is a strict superset, so the shared tools have identical argument shapes and configs are interchangeable.
check_actionDECIDE / GOVERNstdio + remoteGovern an action before it runs → signed approved | denied | escalated + audit id.
POST /api/authorizeclaim_approvalGOVERNstdio + remoteAfter a human approves an escalated authorization, claim its one-use exact-action capability. This is the only exit from an escalated action.
POST /api/authorize/approval/claimscan_agentDEFENDstdio + remoteRed-team an agent's prompt + tools against the attack corpus.
POST /api/redteam/agent-scanverify_proofPROVEstdio + remoteCheck a signed decision envelope's integrity with the OmegaEngine API (server-side HMAC verification).
POST /api/decision/verifyjudgeDECIDEremote onlyScore a scenario → riskLevel, riskScore, safetyScore, recommendedAction. Pass receipt: true (Pro+) to embed a portable Ed25519-signed OmegaReceipt v1.
POST /api/v2/judgeverify_attestationPROVEremote onlyVerify a published Agent Security Attestation by id — Ed25519 signature + RFC 6962 transparency-log inclusion.
GET /api/verify/{id} (same lib path)get_pricingREADremote onlyThe public machine-readable price book. Read this before quoting a price to a human.
GET /api/pricingget_usageREADremote onlyThe calling key's own usage summary, so an agent can self-throttle before hitting its quota (Developer+).
GET /api/billing/usageAdd to your MCP client config:
{
"mcpServers": {
"omega-engine": {
"command": "npx",
"args": ["-y", "@omegaengine/mcp"],
"env": {
"OMEGA_API_KEY": "your-omegaengine-api-key",
"OMEGA_BASE_URL": "https://omegaengine.ai"
}
}
}
}OMEGA_API_KEY — required for check_action (get a key). scan_agent (sample) and verify_proof work without it.OMEGA_BASE_URL — point at your self-hosted instance (http://localhost:3000) to keep everything on-prem. Defaults to the hosted API.Clients that support remote MCP servers can skip npx and connect straight to POST https://omegaengine.ai/api/mcp.
x-api-key header on every request. MCP OAuth is not available yet — API-key auth only. Unlike stdio, the remote endpoint requires the key for all tools.check_action, claim_approval, scan_agent, verify_proof) have the same argument shapes and backing endpoints, so configs are interchangeable; remote additionally serves judge, verify_attestation, get_pricing, get_usage.{
"mcpServers": {
"omega-engine": {
"url": "https://omegaengine.ai/api/mcp",
"headers": { "x-api-key": "your-omegaengine-api-key" }
}
}
}Smoke-test with curl:
curl -s https://omegaengine.ai/api/mcp \
-H 'content-type: application/json' -H "x-api-key: $OMEGA_API_KEY" \
-d '{"jsonrpc":"2.0","id":1,"method":"tools/list"}'cd packages/omega-mcp
npm install && npm run build
node dist/index.js # speaks MCP over stdioThe MCP server is a thin, open-source client — no decision logic runs locally. The engine, the signed audit trail, and the cryptographic proofs live in OmegaEngine (hosted or self-hosted). The drop-in is free and spreads itself; the verifiable control plane is the product.