Developers often face the need to rewrite tools for LLM frameworks multiple times, as each framework like AI SDK, Genkit, or MCP SDK requires its own specific wrapper. This dependency forces libraries to choose a single framework, limiting their portability and increasing maintenance overhead for developers.
The proposed StandardToolV0 interface aims to solve this by defining a framework-agnostic object that includes a name, description, and schemas. By utilizing Standard Schema and Standard JSON Schema, these tools can handle validation and schema conversion without being tied to a specific runtime, allowing them to be used across different AI providers and tools.
Standard Schema is supported by over 30 libraries, including Zod, Valibot, and ArkType, and is integrated into more than 60 projects. The specification consists solely of TypeScript types, meaning it contains no runtime code. This allows developers to define tools once and reuse them in tests, documentation, or across various AI frameworks with minimal adapter code.
The StandardToolV0 specification is currently frozen, with future changes planned for a V1 version. While the proposal offers a standardized way to handle tool execution and metadata, it currently lacks the broad industry backing that supported the initial adoption of Standard Schema, relying instead on a single maintainer.
