VerityLayer MCP - Integration Guide¶
Fail-closed verify-before-you-act trust gate for AI agents. Five tools - verify_fact, detect_injection, moderate_content, redact_pii, guard_action - that run before agent actions execute. Every verdict ships an Ed25519-signed receipt for audit trails.
Command:
npx -y @veritylayer/mcp· Source: mcp.so · Last seen: July 4, 2026
What It Does¶
VerityLayer acts as a mandatory safety checkpoint between your AI agent and the outside world. Before the agent can act on a fact, send a message, or execute a transaction, VerityLayer runs verification checks and returns a cryptographically signed verdict. If any check fails, the action is blocked (fail-closed). This makes it essential for production AI agent deployments where safety and audit trails matter.
Key Capabilities¶
- verify_fact - Cross-references claims against trusted sources before the agent acts on them
- detect_injection - Scans agent inputs and outputs for prompt injection, data exfiltration, and manipulation
- moderate_content - Content safety checks for agent-generated text before it reaches users
- redact_pii - Strips personally identifiable information from agent outputs before external delivery
- guard_action - Final approval gate - the last check before any agent action executes
Every verdict includes: - Ed25519-signed cryptographic receipt - Timestamp and tool identifier - Verdict (pass/fail) with confidence score - Audit trail for compliance
Installation¶
npx -y @veritylayer/mcp
Hermes / Claude Desktop Configuration¶
{
"mcpServers": {
"veritylayer": {
"command": "npx",
"args": ["-y", "@veritylayer/mcp"],
"env": {
"VERITYLAYER_API_KEY": "your-key"
}
}
}
}
CorpusIQ Use Cases¶
1. Agent Output Sanitization¶
"Before sending this agent-generated email to the client, run redact_pii
and moderate_content on the draft. Only release if both pass."
2. Prompt Injection Defense¶
"Run detect_injection on every user input before it reaches the agent's
context window. Block any flagged inputs."
3. Compliance Audit Trail¶
"Show me all VerityLayer verdicts from the last 24 hours - which actions
passed, which were blocked, and why."
Operator Value¶
| Protection | Without VerityLayer | With VerityLayer |
|---|---|---|
| Prompt injection | No detection - agent acts on malicious input | Injection detected and blocked before context entry |
| PII in outputs | Manual review or none | Automatic redaction with cryptographic proof |
| Content safety | Reactive (find issues after sending) | Proactive (block before delivery) |
| Audit trail | Agent logs only (may be incomplete) | Signed receipts for every safety decision |
Safety Architecture¶
User Input → [detect_injection] → Agent Processing → Agent Output
↓
[verify_fact]
[moderate_content]
[redact_pii]
↓
[guard_action] ← Final gate
↓
Deliver / Block
Every arrow includes an Ed25519-signed receipt. The system is fail-closed - if any check cannot complete (network error, timeout), the action is blocked rather than allowed through unsafely.
Pricing¶
Keyless, pay-per-call via x402 (USDC on Base). No subscription required - pay only for the checks you run. Pricing details at veritylayer.com.
Related Resources¶
- External MCP Server Catalog - full curated catalog
- Compliance & Regulatory MCPs - complementary trust and verification tools
- ChronoVerify - image authenticity verification MCP
← External MCP Catalog | MCP Servers Home →
Guide created July 4, 2026. VerityLayer available via npx -y @veritylayer/mcp. Check mcp.so for latest pricing and availability.
Integration with CorpusIQ¶
CorpusIQ is the connector layer that lets AI assistants read the systems a business already runs on - accounting, payments, CRM, email, files and analytics - and answer with the source named. It is read-only by design: it retrieves data and never writes to, or acts in, a connected system, which is what makes it safe to point at live accounts.
This page is part of the CorpusIQ directory, the reference we maintain for the MCP server and agent-skill ecosystem. Run alongside CorpusIQ, a third-party server like this one can be combined with a company's own business data in a single answer.