Give Grok (consumer + API + Grok Build) access to premium reasoning, capital-aware code/trade review, audited sandbox execution, and persistent cross-session memory — settled in sats or USDC (x402).
Use a pre-funded demo account. No top-up required to test.
ivv_j4w83PoN0jght43XnflXNua6nUWJyXhbjrw6OJ5T9x0
https://api.babyblueviper.com/mcp
Dedicated public demo key (non-withdrawable, capped) for trying the review / reasoning / memory tools — the fastest way to feel the value. Persistent workspaces + sandboxed code execution run real compute, so they're a paid feature: register your own key (free) and top up to use those.
Once connected, try: "Use invinoveritas review on this plan before I implement it" or "Reason deeply about the regime for this trade using invinoveritas".
review tool before irreversible actions (deploys, large trades, config changes). The instruction sticks across every chat.AGENTS.md and connected MCP servers. Perfect for our /review (constitutional second opinion) + /execute (audited sandbox with proofs).This is the real power move we just shipped and verified (V1.1).
execute(use_workspace=true) → your agent gets a real persistent directory at /workspace that survives across calls (git clones, files, installed packages stay).workspace_list, workspace_status, workspace_delete let you inspect and clean without burning paid execute calls.review (artifact_type=plan or code_diff, return_suggestions=true) before any commit, push, trade, or risky action. Our own live H1 Hyperliquid trading bot does exactly this before every entry.Proofs: Full multi-step workspace + review example • Live H1 trading bot using the review gate (short versions also available).
This combination (persistent state + mandatory constitutional review) turns Grok Build from a clever coder into a disciplined, auditable agent that serious work can be trusted to. Code execution + workspaces run real compute — try review free on the demo key, then register + top up your own key to unlock the sandbox.
from xai_sdk import Grok
# or OpenAI-compatible client pointing at https://api.x.ai/v1
tools = [
{"type": "mcp", "server_url": "https://api.babyblueviper.com/mcp", "server_label": "invinoveritas",
"authorization": "Bearer YOUR_KEY_HERE"}
]
xAI handles the MCP connection, discovery, and streaming for you.
Connecting the MCP server is step 1. Installing the Skill is what makes Grok remember to use review before every risky action, control costs, and use memory across conversations.
30-second install:
The Skill contains exact instructions for when to call review (our moat — capital-aware second opinion), cost discipline defaults, and special Grok Build behavior. It applies persistently.
After the Skill is active, just say things like "review this plan before we implement" and Grok will call the tool automatically.
The review gate is the fastest way to feel why agents pay for this — start there, then reasoning and memory. All three work on the demo key:
Want stateful, sandboxed code execution + persistent workspaces? That runs real compute, so it needs your own funded key — register free + top up, then use execute(use_workspace=true) (the workspace introspection tools workspace_list/status/delete stay free).
Advanced: When calling from Grok or any client, you can send header X-Client-Hint: grok (or grok-build). We will use this for surface-specific stats and Warden briefs.
Best setup for professional coding agents right now:
One-time (in Grok Build or your AGENTS.md):
Before any write, git commit, shell mutation, or high-stakes action: 1. Use execute with use_workspace=true for any real development work 2. Call invinoveritas review (artifact_type=code_diff or plan, return_suggestions=true) 3. Only proceed after addressing review feedback 4. Use workspace_list / workspace_status to inspect state without extra cost 5. Store key decisions with memory_store (agent_id = "grok-build-this-project")
Recommended first prompt in Grok Build:
"Analyze this repo using a persistent workspace (use_workspace=true). Then review the proposed changes with invinoveritas review before we implement anything."
This is the exact pattern our own systems use. It gives you stateful development + mandatory second-opinion review before anything dangerous. (Uses execute + workspaces, so it needs your own funded key — review alone works on the demo key.)
Building something real? Registered agents can get a one-time 500-sat "first call on us" grant on their own key (non-withdrawable starter — isolated from the shared demo). It's issued to genuine builders, not auto-handed-out: reply to our outreach, post on the agent board, or email below and we'll send a grant token to redeem via /grant_first_call.
Full guide & pricing → | MCP discovery | Server card: /.well-known/mcp/server-card.json
Questions? Post on the agent board, DM via Nostr, or email contact@agents.babyblueviper.com. This is the same production MCP server that powers our own autonomous fleet (including real paid /reason and /review calls today).