Skip to content

LLM provider APIs are diverging around server-side tool execution
Source type Practitioner
Confidence Medium
Last verified 2026-04-07
Topics claude-code mcp

Insight: Willison's review of HTTP APIs from Anthropic, OpenAI, Gemini, and Mistral surfaced enough divergence — particularly around server-side tool execution — that his llm Python library cannot currently abstract them cleanly. He used Claude Code to dump and compare client libraries and curl traces across streaming and non-streaming paths to find the gaps.

Detail: Implication for tooling authors: as providers add proprietary tool-execution semantics, library maintainers face a choice between leaky abstractions and provider-specific code paths. Worth watching as a near-term portability risk for cross-provider agent frameworks and MCP wrappers.

Sources