GroupButler Connectors — Integration Pack

For platform review teams (Claude Connectors / ChatGPT Apps / MCP registries) and white-label partners · Updated 2026-10-15

Remote MCP endpoint

https://bc.gensi.space/mcp (streamable HTTP; tools/list, tools/call). Health probe: GET https://bc.gensi.space/mcp-health. Registry manifest: /.well-known/server.json. Agent card: /.well-known/agent-card.json.

Tools

ToolInputOutputSide effects
screen_token{"chain": "eth|bsc|polygon|sol", "address": "0x..."}aggregated verdict (OK/CAUTION/HIGH_RISK) + per-provider signals (GoPlus/RugCheck/CertiK), attribution includednone (read-only, idempotent, 1h cache)
screen_group_message{"text": "...", "lang": "auto"}scam/spam category + reason code + confidencenone (read-only)

When the tools should be called (trigger contract)

screen_token — when a user pastes any 0x/contract address and asks "is this safe / a scam / honeypot / rug?", before buys/mints/claims, or when an admin vets a CA. Trigger phrases: "check this CA", "honeypot check", "rug check", "safe to buy?". NOT for price predictions or trading advice.

screen_group_message — when a chat message needs scam classification with audit reason codes (fake-admin wording, drainer links, impersonation).

Auth

Beta tier: unauthenticated, rate-limited (fair use), provider attribution required in downstream UIs. Production tier for connectors: OAuth 2.0 (authorization-code) wrapper is prepared; contact [email protected] for a partner client ID and sandbox token.

Test plan for reviewers

  1. GET /mcp-health → expect {"ok": true}.
  2. tools/list → 2 tools with descriptions including when-to-use contract.
  3. Call screen_token with any verified token CA (e.g. USDT on Ethereum) → expect OK verdict with GoPlus attribution.
  4. Call screen_token with a known honeypot CA from public honeypot lists → expect HIGH_RISK with reason codes.
  5. Call screen_group_message with "Send 0.5 ETH to verify your wallet immediately" → expect scam category + reason code.

Privacy & terms

Privacy Policy · Terms of Service · Data sent to screening tools: the contract address or message text only; verdict metadata retained 90 days; no data sale; AI identity always disclosed.

Evidence

Published red-team methodology and per-family detection rates: 2026 Telegram Crypto Scam Report (186 rounds, 68.4% auto-blocked, 0/15 false positives, weak families disclosed).