Shell Company Detector

Shell Company Detector

Score shell-probability from live government registry data — no aggregator "shell flags", the actual raw signals.

Prerequisite: configure the OpenRegistry MCP server

This skill calls OpenRegistry MCP tools (search_companies, get_company_profile, list_filings, fetch_document, etc). Add the server to your AI client config before invoking:

{
  "mcpServers": {
    "openregistry": { "url": "https://openregistry.sophymarine.com/mcp" }
  }
}

Free anonymous tier, no API key required. Restart your client after adding. If the tool calls below return tool not found, the MCP server isn't wired up — see openregistry.sophymarine.com/docs.

What you get

• Live multi-signal probe of every well-known shell-entity indicator, pulled from the government register at query time.
• Unmodified evidence for every signal: director count, accounts filing history, registered address, employee headcount (from accounts), director multi-company concentration, nominee-address patterns.
• Transparent scoring: a shell-probability score (0-100) with each signal's contribution shown, so a compliance officer can justify or dispute the conclusion.
• Cross-jurisdiction applicable — signals are defined generically; per-jurisdiction availability is surfaced.

Shell signals detected

Example prompts

Is UK company 14123456 a shell? Run all signals and score it.

Screen this Delaware LLC (we have the name, no ID) for shell-company red flags. 
How confident can you be with only a name?

For this list of 20 UK counterparties, flag any with >3 shell signals.

Workflow

Step 1 — Resolve the entity

search_companies + get_company_profile

Step 2 — Signal collection

get_officers({ include_resigned: false })            // Signal 1
list_filings({ category: "accounts", limit: 100 })   // Signal 2, 3
get_company_profile (already have)                    // Signal 4, 9
get_officer_appointments (for each director, GB)     // Signal 6
get_persons_with_significant_control                 // Signal 7
fetch_document on latest accounts → parse employees  // Signal 8

For Signal 5 (shared-address reverse lookup): limited native support — IE CRO supports address search via search_companies(query=<address>); for others, the skill reports "Signal 5 unavailable for jurisdiction X" and advises the user to check the address manually via Companies House or equivalent.

Step 3 — Score

Each signal contributes a weight to a 0-100 shell-probability score:

• Strong signals (1-3): 20 points each
• Multipliers (4-7): 10 points each
• Weak signals (8-10): 5 points each

Cap at 100. Report each contributing signal with its evidence:

Shell probability: 75/100

• ✅ Signal 1 (single director): found 1 officer (John Smith, appointed 2024-01)
• ✅ Signal 2 (zero accounts): incorporated 2023-06, no accounts filed (overdue since 2025-03)
• ✅ Signal 4 (overseas registered office): address in Marshall Islands
• ✅ Signal 6 (nominee director pattern): Smith holds 47 other active appointments (via get_officer_appointments)
• ❌ Signal 7: no PSC filings at all (insufficient data to check PSC=director match)
• ❌ Signal 8: no accounts → no employee data
• Total: 20+20+10+10 = 60. Adjusting upward for missing-PSC anomaly (+15 = 75)

Step 4 — Cross-link for deeper investigation

If shell-probability > 50, suggest follow-ups:

• Walk the ownership chain → Cross-Border UBO Chain Walker
• Check the director's other companies → Director Search & PEP Screening
• Check for phoenix pattern (dissolved predecessor at same address / same director) → Phoenix Company Radar

Provenance & Auditability

Every signal is backed by a specific government record cited inline: the officer roster CURL, the filing history entry, the PSC notification record. A compliance officer reviewing the conclusion can verify every signal at the government source.

Important caveats

• Shell signals ≠ proof of fraud. Many legitimate holding vehicles tick several of these boxes (e.g. a personal investment SPV may be single-director + micro-entity accounts). The score is a screening heuristic, not a verdict.
• Jurisdiction-dependent accuracy: Signal 6 (cross-company director footprint) is reliable on GB (native get_officer_appointments), partial on FR / TW, unavailable elsewhere. Signal 8 (employee count) requires accounts to be filed and parseable (GB iXBRL, KR XBRL — rest may not be parseable from PDF).
• False negatives: a sophisticated shell with nominee directors rotated regularly, filed accounts showing minimal but non-zero activity, and a non-shared address can score <30. Use the skill as ONE layer in a multi-layer AML pipeline, not the sole check.

You might also need

• Walk ownership upstream → Cross-Border UBO Chain Walker
• Check director's full footprint → Director Search & PEP Screening
• Detect phoenix pattern → Phoenix Company Radar

Why the data stays fresh

Every signal is collected from live get_* / list_filings calls at query time. When a new filing changes the signals (new director appointed, accounts now overdue → no longer "zero filed"), the next run of the skill reflects it within the government's publication cycle. No stale score.