Guide
Build B2B lead lists from EU business registers with Claude
A practical walkthrough: ask Claude in plain words, let it pick NACE codes, filter official register data by size, age and region, and export a CSV.
7 min read · updated
Official business registers are the most complete free source of company data in Europe, but each one has its own file format, industry codes and quirks. This guide shows how to turn them into a usable B2B lead list by asking Claude in plain language, with EU Company Leads connected as an MCP server.
What you are working with
EU Company Leads loads the bulk open-data files that national registers publish and normalises them into one index. Every company gets a NACE Rev. 2.1 industry code, a normalised legal form, a region, a founding date and, where the register publishes it, a headcount. Sole traders are removed before anything is stored, so the index holds legal entities only.
Countries covered today: France, Sweden, Belgium, Czech Republic, Finland and Estonia. The index currently holds 5,488,441 companies. See all countries for the per-country pages.
Claude talks to the index through four tools:
| Tool | What it does |
|---|---|
lookup_industry |
Turns plain words ("software", "logiciel", "ohjelmisto") or a national code (NAF 62.10Z, TOL 62100) into NACE Rev. 2.1 codes |
search_companies |
Filters companies and returns compact rows, the total match count and a cursor for the next page |
get_company |
Returns one company in full |
export_companies |
Writes every match to a CSV and returns a download link |
If you have not connected the server yet, follow Connecting an MCP server to Claude first.
Step 1: describe the list you want
Start with the kind of request you would give a research assistant. Name the industry, the country, the size and anything else that defines a good prospect:
Find Swedish IT consultancies founded after 2018 in Stockholm county.
List Czech manufacturers of machinery with 50–249 employees, largest first.
Which Finnish companies in architecture and engineering are registered in Uusimaa?
You do not need to know any codes or field names. Claude reads the tool descriptions and picks the parameters itself. What helps is being explicit about the things that change the list: the country, a size range, an age range and a place.
Step 2: industry words become NACE codes
The registers do not store words like "SaaS" or "biotech". They store activity codes. So before searching, Claude calls lookup_industry, which ranks matching NACE Rev. 2.1 codes from official headings and a curated list of everyday terms. For "software" it returns codes such as 62.10 Computer programming activities and 58.29 Other software publishing; for "research" it suggests 72 Scientific research and development.
The search tools only accept exact codes, which is deliberate: guessed codes are the most common reason a list silently misses companies. Codes work at three levels:
- a division (two digits, e.g.
62) covers every class under it; - a group (three digits, e.g.
72.1) covers its classes; - a class (four digits, e.g.
62.10) is the most precise.
Use divisions for broad briefs ("all IT services") and classes when the brief is narrow. Several codes can be combined in one search. If you want to understand the coding system itself, read What is a NACE code? or browse the industry code pages.
Step 3: filter with search_companies
search_companies takes the codes plus any combination of these filters:
| Filter | Notes |
|---|---|
country / countries |
ISO codes, e.g. "FR" or ["FI","EE"] |
nace_codes |
From lookup_industry; divisions and groups include their classes |
include_secondary_nace |
Also match secondary activity codes (default off) |
employees_min / employees_max |
Size bands match if they overlap the range |
include_unknown_size |
Keep companies without employee data when filtering by size |
founded_after / founded_before |
YYYY-MM-DD, inclusive |
regions / cities |
Region names or ISO codes ("Uusimaa", "FR-IDF"), city names |
legal_forms |
Normalised: limited, public_limited, partnership, cooperative, foundation, association, foreign_branch, public_sector, housing, other |
active_only |
Default true: excludes bankrupt, liquidating, dormant and dissolved companies |
name_contains |
Accent-insensitive substring of the name |
sort |
name, employees_desc, founded_desc, founded_asc |
Results come back as compact rows (id, name, NACE code, city, employees, founding date) with the total number of matches. The total matters: it tells you whether your brief describes 40 companies or 40,000, and Claude can suggest tightening or loosening filters before you export anything.
Paging is handled with a cursor. Each response includes a next_cursor when more rows exist; Claude passes it back with identical filters to get the next page (up to 100 rows per page). With verbose=true each row includes address, website and every stored field.
Unknown parameters are rejected with a list of valid ones, so a misspelt filter never quietly returns the wrong list.
Step 4: check individual companies
When a row looks promising, get_company returns the full record: legal form, primary and secondary NACE codes, employees, founding date, address, website and status. Pass the id from the search results, for example FI:0112038-9, or a country plus national ID.
For France, Finland and the Czech Republic, refresh=true re-reads the company live from the national register before answering (INSEE's recherche-entreprises service, PRH's YTJ API and ARES respectively). It is slower, so use it for a shortlist rather than a whole list. Sweden, Belgium and Estonia have no keyless live service, so their records reflect the latest bulk file.
Step 5: export the full list
Conversations are a poor place for 2,000 rows. When you want the list, ask Claude to export it. export_companies takes exactly the same filters, writes every match to a CSV (default 1,000 rows, up to 10,000 per export) and returns a signed download link that is valid for 24 hours. The CSV includes id, name, country, national ID, legal form, NACE code and title, employee range, founding date, status, address, postcode, city, region, website, phone and e-mail where the register publishes them.
If your list is larger than 10,000 rows, split it: by country, by region or by founding-date range. Logged-in users can run the same searches and exports in the web app.
Country caveats that change your list
The registers do not all publish the same fields, and a filter can only use what is there.
| Country | Employee data | What to know |
|---|---|---|
| France | Headcount bands | Only a filtered subset of Sirene is loaded: active legal entities with at least 6 employees, or any size in sections C, J, K or N. There is no band boundary at 5. |
| Sweden | None | Companies that opted out of advertising are not stored. Region is derived from the postcode. |
| Belgium | None | KBO/BCE publishes no headcount. |
| Czech Republic | Bands, about 39% coverage | Some companies only have a 1–3 digit code. |
| Finland | None | PRH does not publish headcount. |
| Estonia | Exact count, about 53% coverage | From the Tax and Customs Board's quarterly file. |
The practical consequence: a size filter on Sweden, Finland or Belgium returns nothing unless you set include_unknown_size=true, and in the Czech Republic and Estonia it drops every company without a reported size. Search responses include notes about this, and Claude will usually relay them. If size matters, filter in the countries that have it and use industry, age, legal form and region elsewhere. The country guides go into each register in detail.
Attribution and responsible use
Every tool response includes an attribution array naming each source and its licence, for example Licence Ouverte 2.0 for INSEE Sirene or CC BY 4.0 for PRH. Keep it with the data when you pass a list on. The full list is on the sources and licences page.
Most of the data describes legal persons and is not personal data, but some fields can still point to a person, for example a small company's registered e-mail address. If you use a list for outreach, you are the controller for that outreach. Using register data for prospecting under GDPR covers what that means in practice.
Example: from brief to CSV
A typical session looks like this:
- You ask: "Find French software companies with 10–50 employees in Île-de-France, newest first."
- Claude calls
lookup_industrywith{"query": "software", "country": "FR"}and keeps58.29and62.10. - Claude calls
search_companieswith those codes,employees_min: 10,employees_max: 50,regions: ["Île-de-France"]andsort: "founded_desc", and reports the total. - You narrow it: "Only SAS and SA, founded after 2015." Claude adds
legal_formsandfounded_after. - You ask for the list. Claude calls
export_companiesand gives you the CSV link.
The same pattern works for any segment. Browse the French software segment or the Swedish engineering segment to see what the data looks like before you start, and request access when you want to run your own searches.