QNSP

Build Pattern · MCP

Build a PQC-Authenticated AI Agent via MCP

Wire Claude or ChatGPT into QNSP via Model Context Protocol — PQC-signed tool calls, audited inference.

Expose QNSP capabilities (vault, KMS, audit, crypto-inventory) to Claude / ChatGPT via Model Context Protocol. Every tool call is PQC-signed and audited; the agent never sees plaintext secrets. Quickstart in under 15 minutes for a Claude Desktop or Cursor integration.

15 minTime to first PQC
MCPPrimary SDK
3Services used

QNSP services used

Stack

Code

Real code, real SDK calls

Snippets reference the published @cuilabs/qnsp SDK surface (TypeScript, Python, Go, Rust mirror byte-for-byte).

Claude Desktop config — add the QNSP MCP serverjson
{
  "mcpServers": {
    "qnsp": {
      "command": "npx",
      "args": ["-y", "@cuilabs/qnsp-mcp"],
      "env": {
        "QNSP_API_KEY": "qnsp_live_..."
      }
    }
  }
}
Programmatic MCP client (Cursor / custom agent)typescript
import { QnspMcpClient } from "@cuilabs/qnsp-mcp";

const mcp = new QnspMcpClient({ apiKey: process.env.QNSP_API_KEY! });

// Every tool call is ML-DSA-signed and audited by QNSP
const tools = await mcp.listTools();
// → [{ name: "qnsp_vault_get", ... }, { name: "qnsp_kms_sign", ... }]

const result = await mcp.callTool("qnsp_vault_get", {
  secretId: "openai-key",
});
// Agent receives a signed envelope; never raw plaintext

Ship it

Get an API key and start building