Connect Claude to government data

This guide goes with the workshop Conecta Claude y ChatGPT a datos del gobierno, which we ran with Lab10 on August 27, 2026. Use it to repeat the workshop step by step, at your own pace.

Goal#

We are going to build a skill in Claude that takes a list of companies, enriches it with each one's registry record and financial statements, and emails the result to your team. Without opening a single portal.

What Croma is#

Croma puts Latin America's public data one call away from your agent. One key opens more than 70 official sources: RUES, Supersociedades, Rama Judicial, SECOP, the Registraduría and many more, all in the same format. You ask Claude about a company and it answers with information pulled from the official portal, with the source and the date. Browse the sources, try the tools or read the documentation.

What a skill is#

A skill is a folder of instructions that Claude loads to do a specific task in a repeatable way. That is how Anthropic defines it. It has three parts: a name, a description that tells Claude when to use it, and the step-by-step instructions.

It is good for two things: defining a process once so you can run it easily next time, and sharing it with your team, who install it from the organization's directory and run it the same way.

A skill describes the process. Croma brings the data, which is why we start by connecting it.

Why Croma exists#

An agent is only as good as the information it can reach. Without data it has nothing to process. With the business registry, the financial statements and the court records at hand, it verifies, cross-checks and decides. In Latin America that information is public, but it is spread across dozens of portals that do not talk to each other. Croma turns it into a tool the agent uses like any other.

Step 1: create your Croma account#

1. Go to platform.usecroma.com/sign-up and create your account. It is free.
2. If it asks you to create an organization, name it after your company and continue.
3. Open platform.usecroma.com/mcp and accept the terms.

Step 2: connect the Croma MCP to Claude#

It works on Claude's free plan. Sign in and click this button:

Add Croma to Claude

Claude opens with the connector already filled in. Click Add, then Connect, and sign in with your Croma account. The tools show up in every new chat.

If you would rather do it by hand, go to Customize → Connectors, click + and then Add custom connector, and paste these two values:

Croma | The API for government data

https://api.croma.run/mcp

To check the connection, open a new chat and ask:

Look up NIT 900843898 in RUES and summarize the company

If Claude answers with Rappi's details, you are set. The per-client guide, ChatGPT included, is at /mcp/claude.

Step 3: the process we are automating#

For each company on the list, the agent makes three calls to Croma:

With that, it builds one row per company and delivers the table as a CSV. The hard part is not the calls, it is the judgment: which record to pick when there are several, what units the figures come in, and what to do when there is no data. That goes in the skill.

Step 4: create the skill in Claude#

Open claude.ai/customize/skills, click Create a skill and paste these three blocks into their fields. The third one goes in the large editor at the bottom, where it says Add file.

Skill name#

lead-enrichment

Description#

Enriches a list of Colombian companies (CSV or text) with public data from RUES and Supersociedades using the Croma tools. Use it when asked to enrich, qualify or prioritize leads, or to research several companies at once.

Instructions#

# Lead enrichment with Colombian public data
 
Enrich every company on the list with its business registry record (RUES)
and its financial statements (Supersociedades) using the Croma tools.
 
## Input
 
A CSV or a list with at least the company name. If it carries a `nit`
column, use it. Otherwise resolve it with the name search.
 
## Croma tools
 
- `rues_entities_by_name`: searches by name and returns candidates with
  their NIT.
- `rues_entity_by_nit`: the full record by NIT, with renewals, legal
  representatives and notices.
- `supersociedades_financial_statements`: annual financial statements by
  NIT, newest first.
 
## Process, one company at a time
 
1. No NIT: call `rues_entities_by_name` and pick the record whose
   `registration_status` is ACTIVA and whose category is SOCIEDAD (not a
   commercial establishment or a natural person). Take its `nit`.
2. With a NIT: call `rues_entity_by_nit`. If the record is not ACTIVA, or
   is a PERSONA NATURAL or an ESTABLECIMIENTO, repeat step 1 with the name
   and keep the active main registration.
3. Call `supersociedades_financial_statements` with the NIT. Use the most
   recent year and the one before. If `reporting_unit` is MILES DE PESOS,
   divide by 1,000 to report in millions of pesos.
4. Compute revenue growth against the previous year and classify: Small
   (under 10,000 million), Medium (10,000 to 100,000 million), Large (over
   100,000 million).
5. From the legal representative take only the name, never the document
   number.
6. If a tool fails or finds nothing, write "no data" in that column and
   move on to the next company. Never invent a figure.
 
## Output
 
First a table, then the same content as a downloadable CSV with these
columns:
 
empresa, nit, razon_social, estado, camara, actividad_ciiu,
ultima_renovacion, representante_legal, anio_reporte, ingresos_millones,
utilidad_neta_millones, activos_millones, crecimiento_ingresos, segmento,
fuente
 
If asked to send the result by email, use Composio's Gmail tool, with the
summary and the full CSV in the body of the message.
 
Close with three lines: which companies came back complete, which came
back partial, and what it would take to complete them.

Click Create. The skill is now available in all your chats.

Step 5: download the sample list#

For the workshop we use three companies everyone knows, with records in both RUES and Supersociedades.

Download lab10-leads.csv

If you would rather build it yourself, this is the whole file:

empresa,nit
Rappi,900843898
Frisby,891408584
D1,900276962

The nit column is optional: without it, the skill resolves the company by name. Frisby is the interesting case. Its NIT leads first to an old registration, and step 2 of the skill corrects it. Details like that are why the skill is worth writing.

Step 6: run the skill#

1. Open a new chat in Claude.
2. Type / and pick lead-enrichment.
3. Attach lab10-leads.csv and write: Enrich these leads.

The first time, Claude asks for permission to use each Croma tool. Accept and let it work. With these three companies the result looks like this (figures in millions of pesos, from the latest reported financial statement):

Three companies take under a minute. A hundred take a coffee.

Step 7: send the result to your team with Composio#

A lead enrichment that stays in the chat is not worth much. The last step is getting it to whoever will call those companies, without leaving Claude.

For that we use Composio, the bridge that connects Gmail to Claude. Composio builds connectors: with one connection, Claude reaches more than 1,000 apps. And unlike the connectors Claude ships with by default, theirs also act: they send the email, create the task, update the CRM.

1. Create your free account at dashboard.composio.dev, find Gmail and click Connect.
2. In Claude go to Customize → Connectors, click + and then Add custom connector, and paste these two values.
3. Click Add, then Connect, and authorize Composio in the browser.

Composio

https://connect.composio.dev/mcp

Back in the chat where the skill ran, ask:

Email the lead enrichment result to name@yourcompany.com: subject "Enriched leads: Rappi, Frisby and D1", a three-line summary and the full CSV in the body of the message.

Claude drafts the email, shows it to you and sends it. Public data, a skill with judgment and the result in your team's inbox, all in a single chat.

What comes next#

Just like we built a lead enrichment here, Croma has more than 70 sources to automate other processes: due diligence with Rama Judicial and the Procuraduría, contract tracking in SECOP, identity checks with the Registraduría, or whatever your team repeats every week. Browse the sources, try the public tools or read the documentation. And for the next workshop, follow Lab10.

Join the community#

If you built the skill, tell us how it went. The Croma community lives on WhatsApp: that is where we share skills, answer questions and announce the next workshops.

Join the Croma community