Adversarial Corporate Opacity MCP

> View on ApifyForge | Use on Apify Store

---

Quick Start

Add to your MCP client (Claude Desktop, Cursor, Windsurf):

{
  "mcpServers": {
    "adversarial-corporate-opacity-mcp": {
      "url": "https://ryanclinton--adversarial-corporate-opacity-mcp.apify.actor/mcp"
    }
  }
}

---

Beneficial ownership detection and corporate opacity analysis via the Model Context Protocol, built for AI agents that investigate entities across 6 international registries and 4 sanctions watchlists. This MCP server implements six distinct anti-concealment algorithms — from BFS ownership graph traversal to Bayesian belief propagation — and delivers a structured opacity score plus formal Enhanced Due Diligence reports that hold up in compliance workflows.

When a corporate structure is deliberately obscured through nominee directors, secrecy jurisdictions, shared registered addresses, or adversarial name variations, standard screening misses it. This server targets exactly those evasion patterns. It orchestrates 15 Apify actors in parallel per tool call, runs cross-lingual transliteration matching against OFAC and Interpol, clusters shell company address farms with DBSCAN, and infers beneficial ownership through loopy belief propagation on a multi-evidence factor graph.

What data can you access?

| Data Point | Source | Example |
|---|---|---|
| 📁 Global corporate registry records | OpenCorporates | 140+ jurisdictions, company names, officers, filing status |
| 📁 UK company filings and PSC persons | UK Companies House | Officers, persons with significant control, filing history |
| 📁 Canadian federal corporations | Canada Corporation Search | Directors, incorporation date, federal status |
| 📁 Australian business numbers | Australia ABN Lookup | Entity type, GST registration, ABN status |
| 📁 New Zealand company registrations | NZ Companies Office | NZBN, directors, registered address |
| 🔗 Legal entity identifiers | GLEIF LEI | Global parent/child corporate relationships |
| ⚠️ US Treasury SDN sanctions | OFAC Sanctions Search | Entity names, aliases, identification numbers |
| ⚠️ Global sanctions and PEPs | OpenSanctions | 100+ programs, politically exposed persons |
| ⚠️ International wanted persons | Interpol Red Notices | Subject profiles, charges, issuing country |
| ⚠️ US federal wanted persons | FBI Most Wanted | Charges, descriptions, known aliases |
| 🌐 Domain registration records | WHOIS Lookup | Registrant, registrar, creation date, nameservers |
| 🌐 DNS configuration records | DNS Record Lookup | A, MX, NS, TXT records revealing shared hosting |
| 🌐 IP geolocation and ASN data | IP Geolocation | ISP, ASN, country, hosting provider |
| 🔒 TLS certificate transparency logs | crt.sh Search | Certificate issuers and shared SSL assets |
| 📍 Geographic coordinates | Nominatim Geocoder | Lat/lon for address clustering analysis |

MCP tools for corporate opacity analysis

| Tool | Price | Algorithm | Best for |
|------|-------|-----------|----------|
| unfold_ownership_graph | $0.045 | BFS with jurisdictional hop penalties | Multi-layered shell structures, nominee director detection |
| screen_with_transliteration | $0.040 | 5-stage phonetic pipeline | Sanctions evasion via name variations, Cyrillic lookalikes |
| detect_registration_bursts | $0.040 | Kleinberg infinite-state automaton | Coordinated shell company creation campaigns |
| cluster_shell_addresses | $0.045 | DBSCAN spatial clustering | Registered agent address farms, co-location detection |
| correlate_infrastructure | $0.040 | Weisfeiler-Lehman graph kernel | Hidden entity relationships via shared domains, IPs, TLS |
| infer_beneficial_owner | $0.050 | Loopy belief propagation | UBO identification from multi-source evidence |
| compute_entity_opacity_score | $0.045 | Weighted composite scoring | Single opacity grade for compliance decisions |
| generate_edd_report | $0.050 | Full 6-algorithm pipeline | Formal EDD/KYC documentation, regulatory filings |

Why use this MCP server for beneficial ownership analysis?

Manual beneficial ownership investigation requires searching each corporate registry individually, cross-referencing sanctions lists by hand, and trying to correlate infrastructure data with ownership records. For a Cayman-registered entity with UK and Canadian subsidiaries, that is 4-6 hours of research before any analysis begins. Standard compliance tools use exact-match or basic fuzzy screening that misses intentional transliteration evasion.

This server automates the entire investigation pipeline in a single tool call:

- Parallel data collection — 3-15 actors run simultaneously per tool call, collapsing hours of research into 30-120 seconds
- Adversarial evasion detection — the 5-stage transliteration pipeline catches Cyrillic lookalikes, diacritic stripping, and name reordering that standard matching misses
- Structured opacity scoring — every entity gets a numeric score with grade (LOW/MEDIUM/HIGH/EXTREME/CRITICAL) and weighted factor breakdown for audit documentation
- AI-native interface — integrates directly with Claude, Cursor, Windsurf, and any MCP-compatible agent
- Pay-per-use pricing — no monthly subscription; a complete 7-tool EDD investigation costs under $0.35

Features

- BFS ownership graph traversal across 6 registries with per-hop opacity penalties: 0.1 for same-jurisdiction hops, 0.3 for cross-jurisdiction, 0.5 for hops through 22 identified secrecy jurisdictions including Cayman Islands (KY), British Virgin Islands (VG), Panama (PA), Jersey (JE), Liechtenstein (LI), and 17 others
- Nominee director detection using 15 formation agent name patterns including Trident Trust, Mossack Fonseca pattern names, Portcullis, Asiaciti, and generic terms like "corporate services", "registered agent", "company formation"
- Circular ownership detection — flags and counts circular structures where entity A owns entity B which owns entity A
- 5-stage transliteration screening — Unicode NFKD normalization with diacritic stripping, Double Metaphone phonetic encoding, Caverphone encoding, Jaro-Winkler distance with prefix bonus, and token-set ratio with phonetic bonus when metaphone codes match
- Kleinberg burst detection — infinite-state automaton using Viterbi-style dynamic programming to find optimal state sequences where high-rate states represent suspicious incorporation bursts; autocorrelation analysis detects periodic registration patterns
- DBSCAN address clustering with epsilon=50m (0.00045 degrees) and minPts=3; co-location suspicion score = entities_in_cluster × (1 − diversity_index) × jurisdiction_risk_weight, where diversity_index is Shannon entropy of entity types divided by log(n)
- Weisfeiler-Lehman graph kernel on DNS/SSL subgraphs: builds infrastructure graphs (domains → IPs → nameservers → SSL issuers), iteratively relabels nodes by hashing neighbor labels over 3 iterations, computes normalized dot product of label histograms; kernel value above 0.7 indicates likely shared control
- Loopy belief propagation on a factor graph with 6 evidence variables: ownership registration, officer overlap, address co-location, infrastructure sharing, sanctions co-occurrence, and temporal co-registration; iterates with damping factor 0.5 up to 50 iterations until convergence
- Weighted composite opacity scoring: ownership depth (25%), transliteration risk (15%), burst anomaly (10%), co-location (15%), infrastructure concealment (15%), beneficial owner uncertainty (20%)
- Formal EDD report generation with severity-graded findings (low/medium/high/critical) and actionable recommendations for compliance file documentation
- Standby mode deployment — MCP server stays warm on Apify for sub-second response initiation

Use cases for beneficial ownership investigation

AML and KYC compliance workflows

Compliance officers at banks, payment processors, and fund administrators need to identify ultimate beneficial owners during customer onboarding. Multi-layered structures through offshore jurisdictions can obscure true ownership through 4-8 corporate layers. unfold_ownership_graph traverses up to depth 10 across all registries, computing opacity penalties at each hop and flagging nominee directors — producing structured evidence that feeds directly into compliance files.

Sanctions evasion detection

Financial institutions screening counterparties face the problem that sanctioned entities deliberately vary their names to evade standard lists. A Russian oligarch's entity might appear as "Ivanов" (Cyrillic О), "Ivánov" (diacritic A), or "Vanovi" (reordered tokens). screen_with_transliteration applies a 5-stage phonetic pipeline across OFAC, OpenSanctions, Interpol, and FBI databases to catch all these variants in a single call, returning a CLEAR/MODERATE/HIGH/CRITICAL severity rating with evidence per match.

Shell company farm identification

Registered agents in Delaware, Wyoming, Cayman, and BVI sometimes host thousands of entities at a single address. cluster_shell_addresses geocodes all registered addresses associated with an entity, runs DBSCAN spatial clustering, and computes a co-location suspicion score per cluster. This surfaces registered agent farms that represent fabricated corporate diversity rather than genuine separate businesses.

Investigative journalism and corporate research

Journalists and researchers investigating offshore financial structures need to connect apparently unrelated entities that share beneficial ownership. correlate_infrastructure maps shared domains, IP addresses, and TLS certificates between entities using Weisfeiler-Lehman graph kernels — finding connections that no corporate registry records. infer_beneficial_owner then combines all available evidence through belief propagation to assign posterior ownership probabilities.

Coordinated incorporation campaign detection

Private equity analysts, regulators, and intelligence teams investigating corporate fraud benefit from detecting when entities were incorporated in coordinated batches — a pattern associated with shell company creation campaigns. detect_registration_bursts applies the Kleinberg automaton to incorporation date sequences and identifies temporal clusters with statistical significance, including autocorrelation analysis to detect periodic (non-random) patterns.

Enhanced Due Diligence documentation

For formal regulatory filings, correspondent banking relationships, or high-value transaction approvals, generate_edd_report runs all six algorithms in sequence and produces a complete report with ownership graph, sanctions findings, burst analysis, address clusters, infrastructure correlations, beneficial ownership inferences, and a composite opacity score with severity-graded findings and recommendations. The JSON output is structured for direct inclusion in compliance audit trails.

How to connect this MCP server for beneficial ownership detection

1. Get your Apify API token — sign up at apify.com, go to Settings > Integrations, and copy your API token.
2. Add the server to your MCP client — paste the configuration below into your client's MCP settings file. Replace YOUR_APIFY_TOKEN with your actual token.
3. Start your agent session — the 8 tools appear automatically in your agent's tool list. No further setup required.
4. Make your first call — ask your agent to investigate an entity: "Use unfold_ownership_graph to map the ownership structure of Meridian Holdings Ltd, jurisdiction KY."

MCP client configuration

Claude Desktop

Add to claude_desktop_config.json:

{
  "mcpServers": {
    "adversarial-corporate-opacity": {
      "url": "https://adversarial-corporate-opacity-mcp.apify.actor/mcp",
      "headers": {
        "Authorization": "Bearer YOUR_APIFY_TOKEN"
      }
    }
  }
}

Cursor / Windsurf / Cline

Add to your MCP settings:

{
  "mcpServers": {
    "adversarial-corporate-opacity": {
      "url": "https://adversarial-corporate-opacity-mcp.apify.actor/mcp",
      "headers": {
        "Authorization": "Bearer YOUR_APIFY_TOKEN"
      }
    }
  }
}

Direct HTTP (cURL)

curl -X POST "https://adversarial-corporate-opacity-mcp.apify.actor/mcp" \
  -H "Content-Type: application/json" \
  -H "Authorization: Bearer YOUR_APIFY_TOKEN" \
  -d '{
    "jsonrpc": "2.0",
    "method": "tools/call",
    "params": {
      "name": "unfold_ownership_graph",
      "arguments": {
        "entity_name": "Meridian Holdings Ltd",
        "jurisdiction": "KY"
      }
    },
    "id": 1
  }'

MCP tool reference

unfold_ownership_graph

BFS traversal of corporate ownership graph across 6 international registries. At each node computes an opacity score from jurisdictional hop penalties, nominee detection (15 formation agent patterns), and circular ownership identification. Prunes at depth 10 or when cumulative opacity exceeds threshold.

Parameters:

| Parameter | Type | Required | Default | Description |
|-----------|------|----------|---------|-------------|
| entity_name | string | Yes | — | Target entity name to trace ownership from |
| jurisdiction | string | No | — | Primary jurisdiction ISO code (e.g. "GB", "KY", "PA") to prioritize a specific registry |
| company_number | string | No | — | Company registration number if known, passed to the relevant registry |
| max_depth | number | No | 10 | Maximum BFS traversal depth (1–10) |

Example call:

{
  "entity_name": "Meridian Offshore Holdings Ltd",
  "jurisdiction": "KY",
  "max_depth": 8
}

---

screen_with_transliteration

5-stage cross-lingual name matching against OFAC, OpenSanctions, Interpol, and FBI watchlists. Catches adversarial transliterations including Cyrillic lookalikes, diacritic evasion, and name token reordering.

Parameters:

| Parameter | Type | Required | Default | Description |
|-----------|------|----------|---------|-------------|
| names | string[] | Yes | — | Entity or person names to screen (1–50 names per call) |
| include_interpol | boolean | No | true | Include Interpol Red Notices in screening |
| include_fbi | boolean | No | true | Include FBI Most Wanted in screening |

Example call:

{
  "names": ["Viktor Petrenko", "Viktоr Petrеnko", "V. Petrenkov"],
  "include_interpol": true,
  "include_fbi": false
}

---

detect_registration_bursts

Applies the Kleinberg infinite-state automaton to incorporation date sequences. Uses Viterbi-style dynamic programming to identify high-rate burst states, then runs autocorrelation to detect periodic (non-random) registration patterns.

Parameters:

| Parameter | Type | Required | Default | Description |
|-----------|------|----------|---------|-------------|
| entity_name | string | Yes | — | Entity or person name whose corporate registrations to analyze |
| jurisdiction | string | No | — | Focus on a specific jurisdiction ISO code |

---

cluster_shell_addresses

DBSCAN spatial clustering (epsilon=50m, minPts=3) on geocoded registered addresses. Computes co-location suspicion score per cluster using Shannon entropy diversity index and jurisdiction risk weighting. Up to 20 addresses are geocoded per call via Nominatim.

Parameters:

| Parameter | Type | Required | Default | Description |
|-----------|------|----------|---------|-------------|
| entity_name | string | Yes | — | Entity or person name to find associated addresses |
| jurisdiction | string | No | — | Focus on a specific jurisdiction |

---

correlate_infrastructure

Builds entity infrastructure graphs (domains → IPs → nameservers → SSL issuers), applies 3-iteration Weisfeiler-Lehman relabeling, and computes normalized dot product of label histograms. Kernel value above 0.7 indicates high probability of shared control.

Parameters:

| Parameter | Type | Required | Default | Description |
|-----------|------|----------|---------|-------------|
| entities | string[] | Yes | — | Entity names to compare infrastructure fingerprints (1–20) |
| domains | string[] | No | — | Known domains in "EntityName:domain.com" format |

Example call:

{
  "entities": ["Meridian Holdings Ltd", "Atlas Capital Partners"],
  "domains": ["Meridian Holdings Ltd:meridian-hld.com", "Atlas Capital Partners:atlas-cap.io"]
}

---

infer_beneficial_owner

Bayesian beneficial ownership inference using loopy belief propagation on a factor graph. Combines 6 evidence types per person-entity pair: ownership registration, officer overlap, address co-location, infrastructure sharing, sanctions co-occurrence, and temporal co-registration. Runs with damping factor 0.5, max 50 iterations.

Parameters:

| Parameter | Type | Required | Default | Description |
|-----------|------|----------|---------|-------------|
| entity_name | string | Yes | — | Target entity to identify beneficial owners of |
| known_persons | string[] | No | — | Known associated persons (directors, shareholders, nominees) |
| jurisdiction | string | No | — | Primary jurisdiction ISO code |

---

compute_entity_opacity_score

Runs all six algorithms and returns a weighted composite opacity score. Component weights: ownership depth 25%, transliteration risk 15%, burst anomaly 10%, co-location 15%, infrastructure concealment 15%, beneficial owner uncertainty 20%. Grades: LOW (<0.15), MODERATE (0.15–0.35), ELEVATED (0.35–0.55), HIGH (0.55–0.75), CRITICAL (>0.75).

Parameters:

| Parameter | Type | Required | Default | Description |
|-----------|------|----------|---------|-------------|
| entity_name | string | Yes | — | Entity to compute opacity score for |
| jurisdiction | string | No | — | Primary jurisdiction ISO code |
| domains | string[] | No | — | Known associated domains for infrastructure analysis |
| known_persons | string[] | No | — | Known associated persons for beneficial owner inference |

---

generate_edd_report

Full Enhanced Due Diligence report combining all six algorithms. Runs all 15 actors across registries, watchlists, and infrastructure sources. Returns structured findings with severity grades (low/medium/high/critical), an overall risk score, and actionable recommendations.

Parameters:

| Parameter | Type | Required | Default | Description |
|-----------|------|----------|---------|-------------|
| entity_name | string | Yes | — | Target entity for Enhanced Due Diligence |
| jurisdiction | string | No | — | Primary jurisdiction ISO code |
| domains | string[] | No | — | Known domains for infrastructure analysis |
| key_persons | string[] | No | — | Key persons to investigate (directors, UBOs) |

Output example

compute_entity_opacity_score response for a Cayman-registered entity:

{
  "entity": "Meridian Offshore Holdings Ltd",
  "opacityScore": {
    "entity": "Meridian Offshore Holdings Ltd",
    "overallOpacity": 0.71,
    "ownershipDepthScore": 0.85,
    "transliterationRisk": 0.20,
    "burstAnomalyScore": 0.60,
    "coLocationScore": 0.75,
    "infraConcealmentScore": 0.55,
    "beneficialOwnerUncertainty": 0.80,
    "grade": "HIGH"
  },
  "dataSources": {
    "corporateRecords": 47,
    "leiRecords": 3,
    "watchlistEntries": 0,
    "geocodedAddresses": 8
  }
}

unfold_ownership_graph summary for the same entity:

{
  "entity": "Meridian Offshore Holdings Ltd",
  "summary": {
    "totalNodes": 14,
    "totalEdges": 13,
    "maxDepthReached": 6,
    "circularOwnership": true,
    "nominees": 3,
    "formationAgents": 2,
    "secrecyHops": 4,
    "totalOpacity": 3.4,
    "riskIndicator": "HIGH"
  }
}

screen_with_transliteration severity result:

{
  "severity": "HIGH",
  "result": {
    "matches": [
      {
        "entityName": "Viktor Petrenko",
        "watchlistName": "Viktor Petrenkо",
        "stage": "double_metaphone",
        "similarity": 0.94,
        "phoneticBonus": 0.08,
        "finalScore": 0.91,
        "source": "OFAC-SDN"
      }
    ],
    "totalScreened": 3,
    "totalMatches": 1,
    "pipelineStats": [
      { "stage": "unicode_normalization", "matchesFound": 0 },
      { "stage": "double_metaphone", "matchesFound": 1 },
      { "stage": "caverphone", "matchesFound": 0 },
      { "stage": "jaro_winkler", "matchesFound": 0 },
      { "stage": "token_set_ratio", "matchesFound": 0 }
    ]
  },
  "watchlistSources": {
    "ofac": true,
    "opensanctions": true,
    "interpol": true,
    "fbi": false
  }
}

Output fields

| Field | Type | Description |
|-------|------|-------------|
| entity | string | The entity name investigated |
| opacityScore.overallOpacity | number | Weighted composite opacity score (0.0–1.0) |
| opacityScore.grade | string | LOW / MODERATE / ELEVATED / HIGH / CRITICAL |
| opacityScore.ownershipDepthScore | number | BFS traversal opacity component (weight: 25%) |
| opacityScore.transliterationRisk | number | Sanctions phonetic match risk component (weight: 15%) |
| opacityScore.burstAnomalyScore | number | Kleinberg burst detection component (weight: 10%) |
| opacityScore.coLocationScore | number | DBSCAN address co-location component (weight: 15%) |
| opacityScore.infraConcealmentScore | number | WL graph kernel concealment component (weight: 15%) |
| opacityScore.beneficialOwnerUncertainty | number | Belief propagation uncertainty component (weight: 20%) |
| summary.totalNodes | number | Entity nodes found in ownership graph |
| summary.totalEdges | number | Ownership relationships discovered |
| summary.circularOwnership | boolean | Whether circular ownership structures exist |
| summary.nominees | number | Nominee directors/officers detected |
| summary.formationAgents | number | Formation agent patterns detected |
| summary.secrecyHops | number | Hops through secrecy jurisdictions |
| summary.riskIndicator | string | STANDARD / ELEVATED / HIGH |
| result.matches[].finalScore | number | Composite transliteration match score (0.0–1.0) |
| result.matches[].stage | string | Pipeline stage that produced the match |
| result.matches[].source | string | Watchlist source (OFAC-SDN, OpenSanctions, Interpol, FBI) |
| severity | string | CLEAR / MODERATE / HIGH / CRITICAL |
| bursts[].burstLevel | number | Kleinberg state level (higher = more anomalous) |
| bursts[].periodicity | number \| null | Detected registration periodicity in days |
| clusters[].suspicionScore | number | Co-location suspicion score per address cluster |
| clusters[].diversityIndex | number | Shannon entropy of entity types in cluster |
| inferences[].posteriorProbability | number | Bayesian posterior probability of beneficial ownership |
| inferences[].converged | boolean | Whether belief propagation converged for this inference |

How much does it cost to run beneficial ownership investigations?

This MCP server uses pay-per-event pricing — you pay per tool call. Platform compute costs are included. Each tool has a fixed price regardless of how many registries or watchlists it queries internally.

| Scenario | Tool | Price | Notes |
|----------|------|-------|-------|
| Quick sanctions screen (10 names) | screen_with_transliteration | $0.040 | OFAC + OpenSanctions + Interpol + FBI |
| Ownership graph traversal | unfold_ownership_graph | $0.045 | Up to 6 registries in parallel |
| Shell address cluster analysis | cluster_shell_addresses | $0.045 | Includes geocoding via Nominatim |
| Beneficial owner inference | infer_beneficial_owner | $0.050 | Full 15-actor evidence collection |
| Full EDD report | generate_edd_report | $0.050 | All 6 algorithms, all 15 actors |
| Complete 7-tool investigation | All tools once | $0.355 | Full anti-concealment analysis |
| Monthly compliance workflow (50 entities) | Mixed tools | ~$10–18 | Varies by tool mix |

Set a maximum spending limit per run in your Apify account to prevent unexpected costs. The server respects the limit and returns a structured error rather than continuing.

Compare this to dedicated KYC/AML platforms at $500–2,000/month with per-query fees on top. At $0.04–0.05 per tool call, most compliance teams spend under $20/month for investigative queries.

How this MCP server works

Phase 1: Parallel data collection

Each tool call dispatches between 3 and 15 Apify actors in parallel using Promise.all. Registry selection is jurisdiction-aware: a GB entity queries UK Companies House plus OpenCorporates plus GLEIF; a KY entity defaults to OpenCorporates plus GLEIF plus all registries in full-scan mode. Actor calls have a 120-second timeout (180 seconds for the full EDD pipeline) with graceful fallback to empty arrays on failure, so partial data always produces a result.

Phase 2: Algorithm execution

Raw registry records, sanctions entries, and infrastructure data feed into six purpose-built algorithms implemented in scoring.ts:

Ownership graph (BFS): Entities become graph nodes. OpenCorporates officer arrays and GLEIF parent-child relationships become edges. Hop penalties are assigned per transition: 0.1 (same jurisdiction), 0.3 (cross-jurisdiction), 0.5 (secrecy jurisdiction from a hardcoded set of 22). Officer names are matched against 15 formation agent patterns using substring search to identify nominees. Circular ownership is detected by tracking visited node IDs during BFS traversal.

Transliteration pipeline: Each input name is processed through 5 stages in sequence. Unicode NFKD normalization strips diacritics and normalizes Cyrillic lookalikes. Double Metaphone and Caverphone produce phonetic encodings compared to watchlist entries. Jaro-Winkler computes character-level similarity with a prefix bonus for names sharing an initial sequence. Token-set ratio computes bag-of-words overlap at the token level, with a phonetic bonus added when metaphone codes align.

Kleinberg burst detection: Incorporation dates from all records are extracted and sorted. The Kleinberg automaton models state transitions where higher states represent higher registration rates. Viterbi dynamic programming finds the optimal state sequence. Autocorrelation is computed at lags 1–30 to detect periodic patterns.

DBSCAN clustering: Address strings are compared using a string similarity function (EPS=0.7 similarity threshold, minPts=2). Clusters are expanded iteratively. Shell score per cluster is computed as entities_per_address × (1 − diversity_fraction). In the full pipeline, actual geocoordinates from Nominatim are used with 50-meter epsilon.

Weisfeiler-Lehman kernel: Each entity's domain portfolio is converted to a subgraph. Nodes are iteratively relabeled by hashing their current label with sorted neighbor labels over 3 iterations. Label histogram vectors are computed per entity and compared via normalized dot product. High kernel values identify entities with structurally similar digital infrastructure graphs.

Loopy belief propagation: Variables represent is_beneficial_owner(person, entity) Boolean states. Factor nodes combine 6 evidence potentials. Messages are passed between variable and factor nodes iteratively with a 0.5 damping factor. Convergence is checked at each iteration; the algorithm stops at convergence or 50 iterations. Posterior probabilities above 0.7 are flagged HIGH.

Phase 3: Composite scoring and response assembly

computeOpacityScore takes the six algorithm outputs and applies the weighted formula to produce a single opacity score with a categorical grade. The generate_edd_report tool additionally assembles all six outputs into a unified EDDReport structure with severity-graded findings and natural-language recommendations.

Tips for best results

1. Provide the jurisdiction code when known. Specifying "KY" routes queries to OpenCorporates + GLEIF only (faster, lower internal cost); omitting it triggers a full scan of all 6 registries (more thorough but 2× slower).

2. Supply known domains to correlate_infrastructure and compute_entity_opacity_score. Without domains, infrastructure analysis returns minimal results. Format as "EntityName:domain.com" — multiple domains per entity are supported.

3. Use screen_with_transliteration before infer_beneficial_owner. Sanctions hits update the sanctions_co_occurrence evidence factor in belief propagation. Running screening first and passing those names to the inference tool gives higher-quality posterior probabilities.

4. For large-scale batch investigations, call tools in parallel. The Apify API supports concurrent runs. Screening 50 entities one at a time takes 50 × 60 seconds; running 10 in parallel reduces wall-clock time to 300 seconds.

5. Start with compute_entity_opacity_score for triage. It runs all algorithms and returns a single grade. Route only HIGH and CRITICAL entities to the more expensive generate_edd_report for full documentation.

6. Set a spending limit on your Apify account. Tools like generate_edd_report trigger up to 15 sub-actor runs internally. A spending cap ensures that unexpected entity complexity (many addresses, many officers) does not result in runaway costs.

7. Use detect_registration_bursts on known persons, not just entities. A nominee director who appears as an officer on 40 companies registered in the same 3-month window is a strong shell company farm signal regardless of entity names.

Combine with other Apify actors

| Actor | How to combine |
|-------|---------------|
| Counterparty Due Diligence MCP | Run broad KYB screening first, then escalate HIGH-risk entities to this server for deep anti-concealment analysis |
| OFAC Sanctions Search | Direct OFAC lookups when you need raw SDN records without the transliteration pipeline overhead |
| OpenSanctions Search | Query the full OpenSanctions dataset directly for custom PEP or watchlist workflows |
| UK Companies House | Pull UK PSC (persons with significant control) records directly for UK entity investigations |
| OpenCorporates Search | Raw registry dat