Agent Adapter

Agent Adapter is an open-source agent framework that lets you write a tool once as a plain function and run it anywhere, safely. Key features: • Define a tool once and use it in LangChain, CrewAI, Google ADK, OpenAI Agents and 16 more frameworks • Call the same tools from TypeScript, Java, C#, Go and 5 more languages • Approvals, budgets, cost tracking and traces on every call • One executor enforcing safety checks across all your agents • One-command skill install for Claude Code, Cursor, Codex, Copilot and 60+ coding agents • Free and Apache-2.0 licensed. The core idea is separation of concerns. Your tool code stays small and framework-agnostic, while a single execution layer handles governance. Each call passes through the executor, which can require human approval before a sensitive action runs, enforce spending limits, record what the call cost, and log a trace for debugging and audit. Changing frameworks means changing one line rather than rewriting your tools, so teams can experiment with different orchestration libraries without losing their safety controls. Adding the skill to your coding assistant gives it the context to scaffold tools and wire them through the executor correctly from the start. Built for enterprises that need auditable agent actions, startups that want to ship quickly with guardrails, and developers who prefer building with plain Python or other supported languages. Teams use it to connect internal APIs, databases and third-party services to multiple agent frameworks, to set per-agent budgets for cost control, and to review traces when something unexpected happens in production.