Cryptographic AI governance for any MCP-aware host. Drop signed, hash-chained, independently verifiable receipts into Claude Desktop, Cursor, Cline, or any other Model Context Protocol client with a single config entry.

Built by WORLD999_LABS. Also known as VeriDOCS (legacy scope name on npm).

npx -y @veridocs/mcp@1.1.0

What this is

@veridocs/mcp is the official MCP server for Proof Layer — the cryptographic governance layer for AI agents. Once installed, it gives any MCP-aware AI assistant the ability to:

Submit agent action proposals for governance evaluation before they execute
Receive signed EXECUTE / BLOCK / REVIEW / SHADOW verdicts
Retrieve and cryptographically verify receipts on demand
Create and operate under scoped, time-bounded delegation plans

Every receipt is Ed25519-signed, hash-chained, and independently verifiable offline using only the published public key at /v1/public-key.

Quick start — Claude Desktop

Add this block to ~/Library/Application Support/Claude/claude_desktop_config.json (macOS) or %APPDATA%\Claude\claude_desktop_config.json (Windows):

{
  "mcpServers": {
    "veridocs": {
      "command": "npx",
      "args": ["-y", "@veridocs/mcp@1.1.0"],
      "env": {
        "VERIDOCS_API_KEY": "vd_live_...",
        "VERIDOCS_API_URL": "https://prooflayer.world999labs.com"
      }
    }
  }
}

Restart Claude Desktop. Ask: "List my recent VeriDOCS receipts." — Claude will call vd_list_receipts and return your audit chain.

Get an API key at prooflayer.world999labs.com.

Cursor / Cline / other MCP hosts

The same command / args / env block works in any MCP-aware host. Place it in the host's MCP server configuration file under whatever key the host uses (e.g. mcpServers in Cursor).

Environment variables

Variable	Required	Description
`VERIDOCS_API_KEY`	yes	Your `vd_live_...` key. Get one at prooflayer.world999labs.com.
`VERIDOCS_API_URL`	no	Defaults to `https://prooflayer.world999labs.com`. Override for self-hosted or staging environments.
`VERIDOCS_TELEMETRY`	no	Set to `off` to disable the anonymous install/heartbeat ping. See Telemetry & privacy below.

Telemetry & privacy

This package sends one anonymous install ping on startup, plus periodic anonymous heartbeats while running. The ping contains:

A locally-generated random install ID (cannot be linked back to you)
The package version (e.g. 1.0.2)
The MCP host name (e.g. claude-desktop, cursor, cline)
Your operating system family (e.g. darwin, linux, win32)
Approximate country code derived from your IP (e.g. US, DE)

We do not collect: your IP address (only the derived country, then discarded), your name, your email, your prompts, your receipts, your API key, your file paths, your machine name, or any other environment variables.

To disable, add "VERIDOCS_TELEMETRY": "off" to the env block in your MCP host config. The flag is checked on every startup.

Raw pings are retained for 90 days then aggregated and deleted. Full policy: PRIVACY.md.

The @veridocs/verify package sends nothing — it runs entirely offline.

Tools exposed

The server registers eight tools:

Tool	What it does
`vd_evaluate`	Submit an action proposal — returns a signed receipt with verdict `EXECUTE` / `BLOCK` / `REVIEW` / `SHADOW`.
`vd_get_receipt`	Fetch a single receipt by its UUID.
`vd_verify_receipt`	Cryptographically verify a receipt's Ed25519 signature and key status.
`vd_list_receipts`	List recent receipts; filter by `agent_id` and `verdict`.
`vd_usage`	Current quota usage, credit balance, billing cycle, and plan tier.
`vd_list_policy_packs`	List all governance policy packs with the active one flagged.
`vd_create_plan`	Human approves a scoped, time-bounded plan covering future agent actions.
`vd_delegate`	Agent requests authorization to perform a specific action under a plan.

Full schema for each tool is exposed via ListToolsRequestSchema — your MCP host will discover them automatically.

How it fits

Your AI assistant (Claude Desktop / Cursor / Cline)
        ↓ MCP (stdio)
@veridocs/mcp                          ← this package
        ↓ HTTPS
Proof Layer kernel
        ↓
Signed receipt → hash chain → verifiable offline

Before any consequential action — sending email, modifying data, executing payment, deleting records — the assistant calls vd_evaluate. Proof Layer evaluates against your active policy pack and returns a verdict plus a signed receipt. The assistant proceeds only when the verdict is EXECUTE.

Verification

Receipts are independently verifiable using only Ed25519 standard library calls and the public key at https://prooflayer.world999labs.com/v1/public-key. No SDK or account required. See the main Proof Layer repo for the full verification protocol and offline tooling.