How do tools work in LangChain agents?

A Tool in LangChain is a callable that an agent can invoke when it needs to interact with the outside world. Every tool has three required attributes: a name (how the LLM refers to it), a description (what it does and when to use it — the LLM reads this to decide), and an input schema (the parameters it expects).

When the agent decides to call a tool, AgentExecutor:

1. Finds the tool by name in its tools list
2. Parses the agent's action into the tool's input format
3. Calls tool.run(input) or tool.arun(input)
4. Returns the result as an "Observation" back to the agent

LangChain ships dozens of pre-built tools in langchain-community: web search (Tavily, SerpAPI), code execution (PythonREPL), database query (SQLDatabase), Wikipedia, file I/O, and more. You access them as:

from langchain_community.tools.tavily_search import TavilySearchResults
from langchain.tools import WikipediaQueryRun

search = TavilySearchResults(max_results=3)
wiki = WikipediaQueryRun()

tools = [search, wiki]

A critical practical point: the tool description matters more than the implementation. The LLM decides whether to call a tool based entirely on reading its description. A vague description leads to incorrect tool selection; a precise description improves agent accuracy.