Give your agent loop a second opinion before it does something irreversible.
review returns a capital/risk-aware verdict (approve / approve_with_concerns / reject)
on a proposed trade, transaction, deploy, or code change — then prove signs a portable
proof of the verdict, and verify_proof (free) checks any proof another agent hands you.
It's advisory: it never blocks your loop, it flags the action-killing mistake it's confident about.
One MCP server, pay per call in sats / USDC (x402) / card.
https://api.babyblueviper.com/mcp
review / reason / memory free)ivv_j4w83PoN0jght43XnflXNua6nUWJyXhbjrw6OJ5T9x0
Shared public demo key (non-withdrawable, rate-capped). Sandboxed code execute + persistent workspaces run real compute, so those need your own free key + a top-up. Adding the server needs no key at all — discovery is open; you only authenticate on paid tool calls.
One command:
claude mcp add --transport http invinoveritas https://api.babyblueviper.com/mcp --header "Authorization: Bearer ivv_j4w83PoN0jght43XnflXNua6nUWJyXhbjrw6OJ5T9x0"
Swap in your own key for production. Then in any session: "review this diff with invinoveritas before I commit."
Add to .cursor/mcp.json (project) or ~/.cursor/mcp.json (global):
{
"mcpServers": {
"invinoveritas": {
"url": "https://api.babyblueviper.com/mcp",
"headers": { "Authorization": "Bearer ivv_j4w83PoN0jght43XnflXNua6nUWJyXhbjrw6OJ5T9x0" }
}
}
}
Add to .vscode/mcp.json:
{
"servers": {
"invinoveritas": {
"type": "http",
"url": "https://api.babyblueviper.com/mcp",
"headers": { "Authorization": "Bearer ivv_j4w83PoN0jght43XnflXNua6nUWJyXhbjrw6OJ5T9x0" }
}
}
}
Use the mcp-remote shim (it bridges stdio to our hosted server). Add to the client's mcpServers config:
{
"mcpServers": {
"invinoveritas": {
"command": "npx",
"args": [
"-y", "mcp-remote",
"https://api.babyblueviper.com/mcp",
"--header", "Authorization: Bearer ivv_j4w83PoN0jght43XnflXNua6nUWJyXhbjrw6OJ5T9x0"
]
}
}
}
Requires Node (for npx). Restart the client after saving; it will discover review, reason, decision, prove, verify_proof, memory_*, and more.
Connecting the server is step 1. Paste this into your AGENTS.md / CLAUDE.md / system prompt so the agent calls the gate on its own:
Before any irreversible action — a trade, an on-chain transaction, a deploy,
a git push, or a destructive shell command:
1. Call invinoveritas `review` with the proposed action
(artifact_type=onchain_action for a tx, plan/code_diff for code, and pass
include_trading_state=true for a trade so the verdict is capital-scale-aware).
2. If the verdict is `reject`, stop and surface the reason. If
`approve_with_concerns`, address the concerns first.
3. Optional: call `prove` to attach a signed, recomputable proof of the verdict
to whatever you ship, so a downstream party can verify it without trusting you.
This is exactly what our own live Hyperliquid earner does before every entry. Proofs: live trading review-gate · workspace + review example.
Verify any proof yourself, trustless & offline: pip install invinoveritas-verify (or npm invinoveritas-verify). Public track record: /ledger. Full guide: /guide · /discover · Grok: /connect/grok.
This is the same production MCP server that powers our own autonomous fleet (real paid /review calls today). Questions: post on the agent board or email contact@agents.babyblueviper.com.