How to define a tool in LangChain

An agent can choose application actions only when each action has a clear name, typed inputs, and a result it can pass back to the model. LangChain represents that contract as a tool, keeping the callable behavior in Python while exposing only the fields the model needs.

The @tool decorator converts a function into a BaseTool object. A Pydantic model supplies field descriptions, defaults, and allowed values for the model-facing schema, while the decorator description explains when the action applies.

A direct invocation can validate the tool contract without an API key or agent loop. The completed program prints the exposed name and schema details, then calls the tool with a support-ticket payload so the returned value depends on the supplied inputs.

1. Open an activated Python project environment that already contains LangChain.

   Related: How to create a virtual environment for LangChain
   Related: How to install LangChain with pip

2. Start define_langchain_tool.py with the imports and Pydantic input schema.

   define_langchain_tool.py

   from typing import Literal
    
   from langchain.tools import tool
   from pydantic import BaseModel, Field
    
    
   class TicketSummaryInput(BaseModel):
       """Input fields for a support-ticket summary."""
    
       title: str = Field(description="Short title from the support queue")
       priority: Literal["low", "normal", "high"] = Field(
           default="normal",
           description="Escalation priority for the routing summary",
       )
       requester: str = Field(description="Name or team that opened the ticket")

3. Add the decorated summarize_ticket() function below the input schema.

   @tool(
       "summarize_ticket",
       args_schema=TicketSummaryInput,
       description="Create a routing summary for a support ticket.",
   )
   def summarize_ticket(
       title: str,
       requester: str,
       priority: str = "normal",
   ) -> str:
       return f"[{priority}] {title} - requester: {requester}"

   snake_case tool names have broad provider compatibility. Side effects such as ticket updates or database writes still need narrow input schemas and application-level authorization checks.

4. Append schema inspection and direct invocation below the tool function.

   schema = summarize_ticket.tool_call_schema.model_json_schema()
   print(f"Tool name: {summarize_ticket.name}")
   print(f"Required inputs: {', '.join(schema['required'])}")
   print(
       "Priority choices: "
       + ", ".join(schema["properties"]["priority"]["enum"])
   )
    
   result = summarize_ticket.invoke(
       {
           "title": "Database backup failed",
           "priority": "high",
           "requester": "Platform Operations",
       }
   )
   print(f"Tool result: {result}")

5. Review the completed define_langchain_tool.py file before execution.

   define_langchain_tool.py

   from typing import Literal
    
   from langchain.tools import tool
   from pydantic import BaseModel, Field
    
    
   class TicketSummaryInput(BaseModel):
       """Input fields for a support-ticket summary."""
    
       title: str = Field(description="Short title from the support queue")
       priority: Literal["low", "normal", "high"] = Field(
           default="normal",
           description="Escalation priority for the routing summary",
       )
       requester: str = Field(description="Name or team that opened the ticket")
    
    
   @tool(
       "summarize_ticket",
       args_schema=TicketSummaryInput,
       description="Create a routing summary for a support ticket.",
   )
   def summarize_ticket(
       title: str,
       requester: str,
       priority: str = "normal",
   ) -> str:
       return f"[{priority}] {title} - requester: {requester}"
    
    
   schema = summarize_ticket.tool_call_schema.model_json_schema()
   print(f"Tool name: {summarize_ticket.name}")
   print(f"Required inputs: {', '.join(schema['required'])}")
   print(
       "Priority choices: "
       + ", ".join(schema["properties"]["priority"]["enum"])
   )
    
   result = summarize_ticket.invoke(
       {
           "title": "Database backup failed",
           "priority": "high",
           "requester": "Platform Operations",
       }
   )
   print(f"Tool result: {result}")

6. Run define_langchain_tool.py to confirm the exposed schema and callable result.

   $ python3 define_langchain_tool.py
   Tool name: summarize_ticket
   Required inputs: title, requester
   Priority choices: low, normal, high
   Tool result: [high] Database backup failed - requester: Platform Operations