← Documentation

EVIDIQ Sentinel

Security preflight for autonomous agents. Scan an MCP endpoint, manifest, Agent Skill, or bundle before your agent connects, authorizes, or pays.

Quickstart

Add the remote Streamable HTTP MCP endpoint to any MCP-capable agent, then use a free validation tool before requesting a paid scan.

claude mcp add --transport http evidiq-sentinel https://mcp.evidiq.dev/sentinel/mcp

Or probe the live pricing discovery endpoint:

curl -s https://mcp.evidiq.dev/sentinel/x402

Prefer a Skill file? Fetch the agent-readable EVIDIQ Sentinel Skill:

curl -s https://mcp.evidiq.dev/sentinel/skill.md

What Sentinel scans

Eight MCP tools

Four free tools help agents validate a target and inspect cost. Four scanners are priced at 0.02 USDT0 per call on X Layer.

Paid scanners

• scan_mcp_endpoint — Inspect a public MCP endpoint without invoking its tools. (x402-paid, 0.02 USDT0)
• scan_mcp_manifest — Analyze supplied tool, prompt, and resource metadata. (x402-paid, 0.02 USDT0)
• scan_agent_skill — Inspect Agent Skill Markdown or a public HTTPS Skill URL. (x402-paid, 0.02 USDT0)
• scan_bundle — Find collisions, shadowing, and toxic flows across targets. (x402-paid, 0.02 USDT0)

Free preflight and verification

• sentinel_capabilities — Read policy, version, limits, and pricing. (free)
• validate_scan_target — Validate a target shape or URL before a paid call. (free)
• estimate_cost — Return the exact atomic and human-readable price. (free)
• verify_scan_report — Recompute the report hash and verify its EIP-191 signature. (free)

Scan an MCP endpoint

First validate the target. Then call scan_mcp_endpoint through your MCP client; this is a paid call and returns an x402 v2 payment challenge until your client provides valid payment authorization.

{
  "name": "scan_mcp_endpoint",
  "arguments": {
    "url": "https://example-agent.dev/mcp"
  }
}

Read the SentinelReport

Sentinel applies deterministic security rules for prompt injection, tool poisoning, shadowing, exfiltration, sensitive data, dangerous capabilities, obfuscation, transport, capability mismatch, and payment-contract compliance. Semantic analysis provides additional context but does not override a high- or critical-severity deterministic finding.

{
  "reportId": "sentinel_…",
  "verdict": "REVIEW",
  "securityScore": 72,
  "findings": [{
    "ruleId": "TOOL_SHADOWING_001",
    "severity": "medium",
    "remediation": "Use a unique tool name and document its scope."
  }],
  "reportHash": "0x…",
  "signature": "0x…",
  "signer": "0x…",
  "storageRoot": "0x…",
  "storageTx": "0x…"
}

Every report is EIP-191 signed. When an anchor succeeds, its 0G Storage root and transaction are included; a storage failure is reported as non-fatal metadata rather than silently changing the verdict.

Verifiable analysis

1. Input is normalized, size-limited, and sanitized before inspection.
2. Deterministic policy rules generate security findings and the final score/verdict.
3. 0G Compute adds TEE-backed GLM-5.2 semantic context without reducing static findings.
4. The canonical report is hashed and signed with EIP-191.
5. The final evidence is anchored to 0G Storage when the best-effort write succeeds.

x402 pricing

Paid scans use x402 v2 with the exact scheme and USDT0 on X Layer (eip155:196). An unpaid request receives HTTP 402 and a PAYMENT-REQUIRED challenge. A compatible client signs the required authorization and retries with PAYMENT-SIGNATURE. Free tools never require a payment header.

Verification and settlement run through the official OKX Onchain OS Payment SDK (@okxweb3/x402-core and @okxweb3/x402-evm). The OKX facilitator verifies each authorization and settles it on X Layer before any paid scan starts. Prices are sent as explicit USDT0 atomic amounts, so the quoted fee token always matches the token registered for the OKX listing.

← Back to docs