How do I validate EU VAT numbers in a Python AI agent?

Steps

1. Install langchain-strale and a chat-model package.
2. Create the toolkit with your Strale API key.
3. Bind vat-validate to your model. vat-format-validate only checks the format and never asks VIES.
4. Ask about a VAT number and run the tool call the model returns.

The solution

# pip install langchain-strale langchain-openai
from langchain_strale import StraleToolkit
from langchain_openai import ChatOpenAI

toolkit = StraleToolkit(api_key="sk_live_YOUR_KEY")
tools = [t for t in toolkit.get_tools() if t.name == "vat-validate"]

llm = ChatOpenAI(model="gpt-4o").bind_tools(tools)
result = llm.invoke("Check if VAT number SE556703748501 is valid and get the company name")

# The model picks the tool and its arguments; your code runs the call.
for call in result.tool_calls:
    tool = next(t for t in tools if t.name == call["name"])
    print(tool.invoke(call["args"]))

How it works

vat-validate asks VIES, the VAT Information Exchange System run by the European Commission, whether a number is registered. It returns whether the number is valid and, where the member state releases them, the registered name and address. Germany, for one, does not release a name or address through VIES.

vat-format-validate is the cheaper, offline check: it confirms the number is well formed for its country and never queries VIES.

Results can be served from a cache of up to 48 hours. If VIES is down, a cached result may come back marked stale: true.

VIES is sometimes slow or unavailable for a single country. Set a timeout, and treat a VIES-unavailable answer as "unknown", not "invalid".

bind_tools only lets the model propose a call. To have the model read the result and answer in words, put the same tools in one of LangChain's agent loops.

Tools in this guide