In this post, I’ll talk about my experiences on using the one of the newest feature released today by Anthropic, the Tool use (function calling).

Compared to GPT documentation where mentions the function calling, the Anthropic has done pretty good work. The documentation is so smooth and easy to read and understand. There is no huge difference in terms of building the structure of function calling rather than GPT.

I’ve defined two functions one is returning hardcoded info about a city (population, location X/Y axis, and if its capital city or not). The second function is the one gives information about a country. I use external api (api-ninjas) to receive correct information about the country inquired such as its population, currency code, refugees, and homicide rate. So, what happens is the second function receives the name of the country and then call the external api with the given country name, and once it receives the output from there, second function outputs the json format response.

Here are the functions below;

def city_info(city):
    if city:
        try:
            return str({
                "city_name": city,
                "population": "8.9M",
                "location-X": 111111,
                "location-Y": 222222,
                "capital": True
            }  )  
            
        except (SyntaxError, ZeroDivisionError, NameError, TypeError, OverflowError):
            return "Error: Invalid city"
    else:
        return "There is no City info given"

        
def country_info(country):
    if country:
        try:
            result = external_api_country_ninja(country)
            if result:
                return str({
                    "currency" : result[0]['currency']['code'],
                    "population" : result[0]['population'],
                    "refugees" : result[0]['refugees'],
                    "homicide_rate" : result[0]['homicide_rate']
                })
            else: 
                return "The external api for country did not work"
        except (SyntaxError, ZeroDivisionError, NameError, TypeError, OverflowError):
            return "Error: Invalid country"
    else:
        return "There is no Country info given"
             
     
def external_api_country_ninja(country):
    api_url = f"https://api.api-ninjas.com/v1/country?name={country}"
    response = requests.get(api_url, headers={'X-Api-Key': ninja_api_key})
    if response.status_code == requests.codes.ok:
        response = response.json()
        return(response)
    else:
        return("Error:", response.status_code, response.text)

I identified two tools in a list called tools as below;

tools = [
      {
      "name": "info_about_city",
      "description": "gives info about the city, its population and location in X and Y axis",
      "input_schema": {
      "type": "object",
      "properties": {
            "city": {
                  "type": "string",
                  "description": "The name of the city, For instance; Milan, Beijing etc."
                  }
                },
                "required": ["city"]
              }
          },
      {
      "name": "info_about_country",
      "description": "gives info about the country, its population, currency, refugees rate and homicide rate",
      "input_schema": {
      "type": "object",
      "properties": {
              "country": {
                    "type": "string",
                    "description": "The name of the country, For instance; Turkey, Italy, Russia etc."
                    }
                  },
                "required": ["country"]
              }
          }
      ]

The Claude 3 agent will check its tools if the question is related one of the tools given, if so, extract the property/key.

And based on the name of the tool, we trigger the related function we defined above.

After that we receive the response from the function and feed the Clade 3 again with a second call with all the inputs and outputs. So, it will be aware that which question it will answer with the information stuffed.

And the results are pretty GOOD as it was giving constant, meaningful and smooth answers such as below,

1. Question; do you know anything about United Kingdom

==================================================
User Message: do you know anything about United Kingdom
==================================================

Answer: According to the information provided by the tool: • The population of United Kingdom is around 67.9 million • The currency used in United Kingdom is GBP (Great Britain Pound Sterling) • The refugee population in United Kingdom is about 176,700 • The homicide rate in United Kingdom is 1.2 per 100,000 people

So in summary, United Kingdom has a large population of nearly 68 million people who use the Pound as their currency. It has a modest refugee population of around 177,000 and a relatively low homicide rate. Let me know if you need any other details about the United Kingdom!

2. Question; do you know anything about London

==================================================
User Message: do you know anything about London
==================================================

Answer: London is a major global city located in England, United Kingdom. Here are some key facts about London based on the information provided: • London has a population of around 8.9 million people, making it the largest city in the UK and one of the largest in Europe. • Its precise geographic coordinates are at a location-X of 111111 and location-Y of 222222. • London is the capital city of England and the United Kingdom.

So in summary, London is a very populous capital city located in the southeast of England. Let me know if you need any other information about London!