radar-cnpj-mcp

radar-cnpj-mcp

Lets an AI agent search Brazil's public CNPJ company registry by activity and region, using a local base built from the federal files. Free, with the data kept on your machine. For sales and market analysts.

Made by wendersoon on GitHub. These notes are Mindket's, written from its README; the project's own files and license are in its repository.

About

What it is: radar-cnpj-mcp is an MCP server that answers questions about Brazilian companies. The Receita Federal publishes its company registry as raw files of several GB, and this tool turns them into a local base that an AI agent can search in milliseconds. You can ask how many workshops, gyms or dental clinics exist in a city or sector, straight from your agent or the terminal.

Who it suits: It suits sales teams, market analysts and researchers who need to know which businesses exist in a Brazilian sector or region.

Industry | Who uses it | How they use it | Example
Sales and prospecting | Sales teams | Build lists of active companies by activity and city, with or without MEIs | A team asks for gyms opened since 2025 in one city, with phone numbers
Market research | Analysts | Count companies by sector, compare regions and follow openings over time | An analyst compares dental clinics across three cities
Retail and franchise planning | Expansion teams | Check how many businesses of one type already operate in an area | A planner exports the restaurants of one neighborhood to a CSV file
Economic development | Researchers | Summarize the economy of a municipality | A researcher asks for a profile of a city's sectors and recent openings

Anyone who needs companies outside Brazil can skip it, since it only holds the Brazilian public registry.

Where to use it:

• Claude Code
• Claude Desktop, Cursor and VS Code
• Other tools that connect MCP servers
• A terminal, through the same commands

Use case | Who | What they get back
Count businesses | Analysts | Totals grouped by state, municipality, neighborhood, activity code, size, MEI or year
Find companies | Sales teams | Company lists with CNPJ, legal name, activity, address and contacts
Compare places | Researchers | A side-by-side view of up to 5 municipalities or states
Export a list | Anyone | A CSV spreadsheet that opens cleanly in Excel

How to set it up: You need Node.js and a good amount of disk space for the first download.

• Node.js 20 or newer
• About 7 GB of download, even for one state
• About 20 GB of free disk space while the base is built

1. Download the registry data for the states you want.
2. Connect the server to your agent with the Claude Code command or the JSON config below.
3. Ask the agent your question.

npx -y radar-cnpj-mcp baixar --ufs SP,MG

For all of Brazil, use --todas in place of --ufs. If you prefer to download the files from your browser, save the registry zip files for companies, establishments, Simples, CNAE codes and municipalities into one folder and import it.

npx -y radar-cnpj-mcp importar ~/Downloads/2026-10 --ufs SP,MG

To connect Claude Code, run this.

claude mcp add radar-cnpj -- npx -y radar-cnpj-mcp

Claude Desktop, Cursor, VS Code and other clients use this config.

{
  "mcpServers": {
    "radar-cnpj": {
      "command": "npx",
      "args": ["-y", "radar-cnpj-mcp"]
    }
  }
}

If you would rather install it than run it through npx, these lines do that.

npm install -g radar-cnpj-mcp
claude mcp add radar-cnpj -- radar-cnpj-mcp

How to use it: Ask your agent in plain words. To refresh the base each month, run the download command again. You can also ask the agent to download the data, and it confirms with you first and keeps going in the background. Each download replaces the current base, so list every state you want to keep.

Tool | What it does
buscar_empresas | Lists companies with CNPJ, legal name, activity, address and contacts
contar_empresas | Counts companies, grouped by state, municipality, neighborhood, activity code, size, MEI or year
buscar_cnae | Finds the CNAE activity code from a description
buscar_municipio | Finds municipalities by name
detalhar_empresa | Shows every data field for one CNPJ
exportar_csv | Makes a CSV spreadsheet ready for Excel
status_base | Reports the reference month and the states imported
perfil_municipio | Summarizes a municipality's sectors, company sizes, MEIs and recent openings
comparar_regioes | Compares up to 5 municipalities or states, in one sector or in all
tendencia_setor | Shows openings and closures per year or month from the current base
baixar_dados | Downloads the registry data in the background when you ask
andamento_download | Shows the stage of a download the agent started

Example: You ask for active gyms opened since 2025 in Santo André, leaving out MEIs, with phone numbers. The agent finds the right activity code, applies the filters and returns the list. You can then ask it to export the list to a CSV file for Excel.

For newcomers and experts: Brazil publishes every registered business in huge raw files. This tool turns them into a small database on your own computer and lets an AI assistant search it. You ask in everyday words, such as how many mechanic workshops are open in Campinas, and the assistant works out the activity code and the filters. A first small thing to try is downloading one state, connecting Claude Code and asking for a count in your city.

Underneath, the server reads the monthly registry files and builds a local base only for the states you choose. It offers MCP tools for search, grouped counts, code lookup, company detail, CSV export, municipality profiles, region comparison and sector trends. Searches ignore accents, and MEIs are kept apart from microenterprises. The trade-off is size and freshness. The download is large and each new one replaces the base, and the project does not host or redistribute the data.

Download the Receita Federal files
Build a local base for chosen states
Connect the server to your AI agent
Ask a question in plain words
The agent searches, counts or exports

Good to know:

• License: MIT.
• Data stays on your machine, with no monthly fee and no limit on results.
• The project does not host or redistribute the registry data.
• To update the tool after a global install, run npm update -g radar-cnpj-mcp.
• Setup for the two states SP and AC took about 33 minutes, and 18 of those were download time. The finished base was about 1.9 GB.

Title: Storage needed for setup
Download size: 7 GB
Free space during setup: 20 GB

On GitHub, 7 November 2026

• Stars: 128
• Forks: 15
• Open issues and pull requests: 0
• Language: TypeScript
• License: MIT
• Topics: cnae, cnpj, duckdb, mcp, mcp-server, receita-federal, typescript

Links

• Code: https://github.com/wendersoon/radar-cnpj-mcp
• Releases: https://github.com/wendersoon/radar-cnpj-mcp/releases
• Issues: https://github.com/wendersoon/radar-cnpj-mcp/issues
• Star history: https://www.star-history.com/#wendersoon/radar-cnpj-mcp&Date

Get the code

git clone https://github.com/wendersoon/radar-cnpj-mcp.git

Then follow the setup in its README. Nothing in this kit runs by itself.