LangChain tool integration boundary

Evidence-backed facts

1. LangChain tools expose callable functions and typed schemas to agents and can receive runtime context, state, or stores. informative
2. The agent chooses a tool call and the LangChain runtime invokes application code, middleware, or a configured tool node. informative
3. Tool context, secrets, stores, external APIs, error handling, and human approval remain separate host-controlled boundaries. informative
4. Tool selection and schema conformance do not prove correct reasoning, authorized mutation, external completion, factual output, or production use. informative

Primary sources