UK Due Diligence MCP Server

Official-source UK due-diligence data for AI agents.

Search Companies House, Charity Commission, The Gazette, HMLR price-paid data and HMRC VAT records, plus screen names against the OFSI, OFAC, EU and UN sanctions lists. Exposes atomic MCP tools for company ownership, officers, cross-company appointment history, secured charges, insolvency notices and related registry evidence — the consuming agent decides how to investigate, not the server.

Every data source is a legally-mandated register with a free official API. Zero paywalls.

---

Data sources

---

Quick start

Hosted (no install)

{
  "mcpServers": {
    "uk-due-diligence": {
      "type": "http",
      "url": "https://uk-due-diligence-mcp.fly.dev/mcp"
    }
  }
}

Local (uvx)

{
  "mcpServers": {
    "uk-due-diligence": {
      "type": "stdio",
      "command": "uvx",
      "args": ["uk-due-diligence-mcp"]
    }
  }
}

See Configuration for the environment variables it needs.

---

Tools

Companies House

Charity Commission

HMLR Land Registry

The Gazette

HMRC / Sanctions

Cross-register

---

Examples

Resolve "TAS Engineering" to a company number and check whether it has any outstanding charges.

→ company_search to find the company number, then company_charges for the full secured-debt picture (not just the has_charges summary).

Has Gareth Davies (director of company 06333469) been connected to any other companies, including ones that no longer exist?

→ company_officers to get his officer_id, then officer_appointments to surface every company he's held an appointment at, current or historic.

Is "MEL Precision Limited" in the middle of insolvency proceedings, and what does that actually mean legally?

→ gazette_insolvency to find the notices, then gazette_notice to read the full legal wording before drawing conclusions from the notice label alone.

Screen "Acme Trading Ltd" and its officers against sanctions lists.

→ company_officers for the officer names, then sanctions_screen against the company name and each officer.

---

Limitations

Things worth knowing before trusting output:

• land_title_search returns Price Paid transactions, not ownership. It does not return current proprietor/title data — HMLR's Price Paid Index only records historic sale transactions.
• Sanctions screening is exact/alias matching, not compliance clearance. A company/entity legal name matches reliably; person names with transliteration variants may not. An empty result is not a guarantee of clearance, and a hit on a common name may need disambiguation.
• has_charges: null means the check could not be confidently completed — a charges-endpoint outage, or a charge with an unrecognized status. Treat it as unresolved, not as "no charges."
• officer_appointments returns historical relationships. Companies House doesn't auto-resign a director when a company enters insolvency — check company_status on each appointment, not just resigned_on.
• Official-source data can be incomplete or delayed. These are the same registers a human would check, with the same latency and gaps.

---

Configuration

HMLR, The Gazette, and the sanctions lists require no credentials.

---

Project structure

uk-due-diligence-mcp/
├── server.py           # FastMCP init, tool/resource registration, transport config
├── companies_house.py  # company_search/profile/officers/psc, officer_appointments, company_charges
├── disqualified.py     # disqualified_search, disqualified_profile
├── charity.py          # charity_search, charity_profile
├── land_registry.py    # land_title_search (SPARQL Price Paid Index)
├── gazette.py          # gazette_insolvency, gazette_notice
├── hmrc_vat.py         # vat_validate (OAuth2 client-credentials)
├── sanctions.py        # sanctions_screen (OFSI/OFAC/EU/UN consolidated lists)
├── search_fetch.py     # search, fetch (cross-register fan-out)
├── models.py           # Pydantic v2 output models
├── http_client.py      # Shared httpx clients, retry backoff, error formatting
├── fly.toml
├── Dockerfile
├── pyproject.toml
└── .env.example

---

Licence

MIT