Claude Code-Execution Call Format

[jmartin1]

Is it possible to use the code execution tool with instances of Claude hosted through Databricks? If I try to format the payload like in Anthropic's documentation, I get an error that the function isn't defined properly: 

    tools=[{
        "type": "code_execution_20250825",
        "name": "code_execution"
    }]

Status code: 400
Response: {"error_code":"INVALID_PARAMETER_VALUE","message":"INVALID_PARAMETER_VALUE: Missing 'function' in the tool specification."}

And if I do define code-execution in the payload according to OpenAI's schema, I get a response with the generated code, but the model doesn't actually run it and return the results in the same output. 

[Louis_Frolio · Accepted solution]

Greeting @jmartin1 , 

The Anthropic “code execution” tool isn’t supported through Databricks’ Foundation Model APIs. Databricks exposes an OpenAI-compatible tools interface, and the only supported tool type today is a function defined with a JSON schema. That’s why you’re seeing the error about a missing function field in your payload. The Databricks REST interface also differs from Anthropic’s API, which means Anthropic’s tool specifications can’t be used directly.

Why you’re seeing that error

Your payload is using Anthropic’s tool format (for example, type: “code_execution_20250825”), but Databricks expects tools to be defined using the structure {“type”: “function”, “function”: { … }}. When the server processes your tools array and doesn’t find a function object, it returns the error: “Missing ‘function’ in the tool specification.”

Cheers, Lou