Tool schemas

What is JSON Schema?

• JSON schema is a data validation specification—a set of rules for validating any JSON data
• Not specific to language models; widely used technology that has existed for many years
• LLM community adopted it as a convenient standard for tool calling

Structure of a Tool Configuration Object

• Name: identifier for the tool (e. g. , "get_weather")
• Description: explains what the tool does, when to use it, and what data it returns
• Best practice: 3–4 sentences long
• Input Schema: the actual JSON schema spec describing function arguments

Input Schema Details

• Specifies each argument the function accepts
• For each argument, define:
• Type: data type (e. g. , string, number)
• Description: explains what the argument controls and how it affects the function
• Best practice: 3–4 sentences long

Practical Workflow for Creating Schemas

• Use Claude to generate JSON schemas automatically
• Prompt Claude with: your tool function + best practices from Anthropic API documentation
• Copy the generated schema into your code
• Naming convention: function_name_schema (e. g. , get_current_datetime_schema)

Type Safety in Python

• Import ToolParam from anthropic. types
• Wrap your schema dictionary with ToolParam() to prevent type errors later
• Not strictly required for code to work, but prevents type checking issues

Takeaways

• JSON Schema is a widely-used data validation standard that the LLM community adopted as a convenient format for tool calling
• A tool configuration requires a name, description (3–4 sentences), and an input schema that specifies each argument's type and purpose
• Use Claude to generate JSON schemas automatically by prompting it with your tool function and Anthropic's best practices
• In Python, wrap schema dictionaries with ToolParam() from anthropic. types to prevent type checking errors
• Argument descriptions should be 3–4 sentences explaining what the argument controls and how it affects the function