Global Database's MCP server is now open to everyone.

Global Database's MCP server is now open to everyone.

From today, any MCP client — Claude, ChatGPT, Cursor, or an agent you built yourself — can query 600M+ companies drawn from 400 government registries across 200+ countries. It takes about two minutes to connect, and it's free to try.

Below: why we built it, four sessions we recorded without editing, where the data actually comes from, and how to connect it. If you'd rather watch than read, the 56-second version is right here.

Four real sessions. Nothing mocked up. · Watch on YouTube

Why it exists

Ask any frontier model to verify a company and watch what happens.

It will give you something. A registration number with the right number of digits. A director who sounds plausible. A turnover figure that lands in the right order of magnitude. Every one of those answers is generated from the shape of the filings it saw in training, not from the filing itself. Sometimes it's right. You have no way to tell which times.

The second failure is quieter and almost as bad: the model refuses. It tells you it can't access live company records and suggests you check the registry yourself. Which you were going to do anyway. So what was the AI for?

Neither failure is a reasoning problem. The model can read a balance sheet perfectly well. It just doesn't have one.

That's a retrieval problem, and retrieval problems have solutions.

The Global Database MCP

We built an MCP server that connects your AI directly to primary, government-sourced company data — 600M+ companies, drawn from 400 government registries across 200+ countries.

Pick a company you know inside out and check our record against what you already know. That's the fastest way to find out whether any of this is real.

Once it's connected, your AI stops guessing. It queries the registry, gets the filed record back, and reasons over the real thing.

What's happening in those four sessions

Company financials

i need the complete company financials for H & T GROUP LIMITED, for the last 5 years and i want you to identify key findings

Back comes FY2020 to FY2024 — turnover, gross profit, EBITDA and margin, EBIT, interest, tax, profit after tax, diluted EPS, headcount. Then the balance sheet: total assets, the pledge and loan book, stock, cash, goodwill, debt, net debt, gearing, current ratio, ROE. Then cash flow, including the working capital movements.

Five years of filed accounts, structured, in one turn. That isn't a summary of a company. It's the accounts.

Corporate hierarchy

identify the full corporate hierarchy tree of JAMES WALKER (LEITH) LIMITED

Back comes the ultimate parent, eight direct subsidiaries, a second tier below those — 19 entities with company numbers attached, sorted into three operating verticals. Then the shareholder register above the top company, summing to 100% across 19,390 ordinary shares, with trusts and individuals named.

Try assembling that by hand. It's an afternoon.

And once the tree exists, you can screen it. Any company or person can be checked against PEP and sanctions lists — a single name, or every entity, officer and owner in the group, in one pass. That's the check that normally forces an analyst to paste 19 names into a screening tool one at a time.

VAT validation

global database, validate the following vat number in uk: GB398297481

A number goes in. A legal entity comes out.

Bulk prospecting

identify 50 companies in italy, with the following codici ateco sic code: 10520 — Manufacture of ice cream and other edible ice

Fifty companies, enriched with website, FY2024 turnover, employee count, phone, VAT number, and a data-quality flag. Combined turnover of roughly EUR 1.75bn across around 3,240 employees. Exported to a spreadsheet.

Same session. Same connection. No CSV export, no vendor portal, no procurement cycle.

What it can do

• Financials — P&L, balance sheet, cash flow, filed accounts across multiple years, public and private
• Corporate hierarchy and linkage — ultimate parent, subsidiaries, full group structure
• Ownership — shareholders and UBOs, with the registry record behind them
• Validation — VAT numbers and company registration numbers across jurisdictions
• Officers — directors, secretaries, appointment history
• PEPs and sanctions screening — check any company or person against PEP and sanctions lists: a single entity, or every company, officer and owner across an entire corporate group in one pass
• List building — filter by SIC, ATECO, NACE, geography, size, revenue, then enrich

Where the data comes from

This is the part that matters, and it's the part most vendors are vague about.

Our data is primary and government-sourced. We take it from 400 government registries — Companies House, the Registro Imprese, each US Secretary of State, and hundreds more. When our record says a company was incorporated on a date, that date came from the registry that incorporated it.

That distinction is invisible right up until the moment it isn't: an audit, a regulator's question, a deal that turns on who actually owns what. Then it's the only thing that matters.

Global Database is ISO/IEC 27001:2022 certified.

Install it

Why we built it this way

There's a bet underneath this, and it's worth stating plainly.

Every serious AI agent is going to end up in the same place: it will be judged on whether its answers hold up. Fluency stopped being a differentiator the moment everyone had it. What separates a useful agent from an expensive one is whether there's a verifiable source under the output.

For company data, that source is the registry. Not a model's recollection of a registry. Not a derived score. The filing.

We spent a decade getting access to 400 of them. The MCP is the shortest path between that work and your agent.

Frequently asked questions