The full upstream README, mirrored here for reference. Install config, tool schemas, adoption signals, and an original overview live on the Invinoveritas listing page.
An independent review gate for AI agents, as a hosted MCP server. Before your agent takes an
irreversible action — shipping code, running a shell command, signing a transaction — it calls
review for a neutral second opinion from a party that isn't the agent, and gets back a
recomputable, Bitcoin-anchored proof anyone can re-verify without trusting the agent or us.
The one thing an agent can't self-serve is a trustworthy verdict on its own output. This is that verdict — model-agnostic, advisory (it never blocks), and checkable from the bytes.
https://api.babyblueviper.com/mcp/verifycurl -X POST https://api.babyblueviper.com/register -d '{"label":"mcp"}'verify_proof and ledger are always free / no auth.| Tool | What it does | Auth |
|---|---|---|
review | Independent verdict (approve / approve_with_concerns / reject) on a code diff, shell command, plan, config, trade, or on-chain transaction, with ranked issues + a portable signed proof. | Bearer (paid) |
verify_proof | Recompute and verify a signed invinoveritas proof someone handed you — confirms it without trusting the presenter or us. | free |
witness | Anchor a third party's exact claim bytes, unmodified and unjudged — pure notarization (receipt + timestamp + integrity), distinct from review (our verdict). Source is self-declared and marked as such in the signed proof. | Bearer (paid) |
Also on this endpoint: ledger (free public verdict track record), validate (backtest reality-check), audit_agent_readiness, ledger_submit, conformance_certify. 8 tools in all; the server never moves money or assets for the user.
Cursor — one-click: Add invinoveritas to Cursor (it signs in with OAuth on the first paid call), or edit ~/.cursor/mcp.json (global) / .cursor/mcp.json (project):
Cline — Add via the MCP marketplace, or in cline_mcp_settings.json:
Gemini CLI — install as an extension:
Claude Code:
Devin — Settings → Connections → MCP servers → Add a custom MCP → HTTP (Streamable), URL
https://api.babyblueviper.com/mcp/verify, Auth Header Authorization: Bearer YOUR_API_KEY.
OpenHands — config.toml:
Any MCP client — point it at https://api.babyblueviper.com/mcp/verify (Streamable HTTP) with header
Authorization: Bearer YOUR_API_KEY. Protocol 2025-06-18; the handshake, tools/list, and
tools/call work out of the box.
Tell your agent: before opening a PR or running an irreversible command, call review with the
diff/command as artifact, set artifact_type (code_diff / shell_command / onchain_action /
general), and sign: true. On reject, hold and fix; on approve, attach the proof so reviewers
can re-verify it.
Fast judgment models such as TypeSafe's Jev (jev_gate in the community
jev-mcp server) screen every tool call in a few hundred
milliseconds and answer allow / confirm / block. They are a good first line. What they don't
give you is an independent verdict someone else can check later. Use both: let the fast gate handle
the routine calls, and escalate only the risky or irreversible ones to invinoveritas.
The fast gate keeps latency and cost low on the many calls; the signed verdict covers the few that matter, and leaves a record that doesn't depend on trusting either the agent or us.
review with sign: true returns a proof that is a BIP-340-signed Nostr event, published to a
public, Bitcoin-anchored ledger. Verify it offline against our published key — recompute the event
id and check the schnorr signature — or POST it to https://api.babyblueviper.com/verify-proof
(free, no auth). Trust the math, not us. Wins and losses are published; the track record is a
forensic artifact, not a score.
pip install invinoveritas-verify / npm i invinoveritas-verifyhttps://api.babyblueviper.com/.well-known/mcp/server-card.jsonMIT