Migration guides

Move existing tool code onto the adapter one step at a time. Each guide has a mapping table, a before-and-after example, what you gain and lose, and a checklist.

flowchart LR
  R["hand-written provider tool calling"] --> REG[("Registry + Executor")]
  L["LangChain @tool"] --> REG
  A["a hand-rolled approval flow"] --> REG
  G["a LangGraph app"] --> REG
  REG --> ANY["any model API, framework,<br/>HTTP or MCP client"]

From	Guide
tools of LangChain, CrewAI, ADK, Pydantic AI and other frameworks, found for you	agent-adapter migrate: a read-only report and a suggested diff
raw OpenAI or Anthropic tool calling (JSON schemas and a dispatch dict)	Raw tool calling
LangChain @tool functions	LangChain tools
a "needs approval" table you wrote yourself	Hand-rolled approval
a LangGraph graph	Migrate from LangGraph (tutorial)

The order that works

1. Register the tools; keep calling them the old way.
2. Route every call through one Executor.
3. Turn on the policies you need: confirm=, budgets, caching.
4. Delete the old schemas, dispatch tables and approval code.

Every step ships on its own and can be rolled back.

The examples are examples/guide_migrate_*.py; tests/docs/test_migration.py runs them.