MCP prompt workflow

Shell Company Detector

Flag 1-director + no-accounts + overseas-office shells using live government-registry signals.

Combines heuristics: 1 officer (often foreign), zero filed accounts, registered-agent address shared with N+ other companies, dormant status, no charges, no establishments. Outputs a 0-100 risk score with contributing signals.

When to use

AML / sanctions screening, tax-risk scoring, 'trade-based money laundering' investigations.

Inputs

ParameterDescription
company_name_or_idTarget company.
jurisdictionISO code.

What the prompt does

Under the hood the prompt orchestrates these tool calls:

  1. get_company_profile
  2. get_officers
  3. get_financials (or list_filings category=accounts)
  4. get_charges
  5. list_establishments (where applicable)
  6. Aggregate signals

How to invoke

On any MCP client that supports prompts (Claude Desktop, Cursor, Cline, Goose, Zed):

  1. Connect to https://openregistry.sophymarine.com/mcp.
  2. Call prompts/list — you'll see shell_company_detector in the array.
  3. Call prompts/get with {"name":"shell_company_detector","arguments":{…}}.
  4. The server returns a pre-built message stream the model can execute directly.

Source skill

This prompt is also distributed as a Claude Agent Skill in the docs repo:

github.com/sophymarine/openregistry/skills/shell-company-detector