LangChain JS agent with a custom tool

Use a LangChain agent when you want a model loop that can choose tools and respond with the result.

This example creates an agent with one custom weather tool.

Install

npm install langchain @langchain/openai zod

Set your API key:

export OPENAI_API_KEY="your_api_key_here"

Code

import { createAgent, tool } from 'langchain'
import { z } from 'zod'

const getWeather = tool(
  async ({ city }) => {
    return `It is always sunny in ${city}.`
  },
  {
    name: 'get_weather',
    description: 'Get the weather for a given city.',
    schema: z.object({
      city: z.string().describe('The city to get weather for.'),
    }),
  }
)

const agent = createAgent({
  model: 'gpt-5.5',
  tools: [getWeather],
})

const result = await agent.invoke({
  messages: [
    {
      role: 'user',
      content: 'What is the weather in San Francisco?',
    },
  ],
})

console.log(result)

What this gives you

The agent receives the user message, decides whether to call get_weather, runs the tool, and returns the model's final response.

Use this pattern when you want:

• A small agent with one or two tools
• A portable model/tool setup
• A path toward middleware, tracing, or LangGraph later

Common mistakes

• Leaving the tool description vague. The agent decides whether to call a tool from its name and description, so a clear description is what makes it fire at the right time.
• Skipping the Zod schema. Without it the model has no contract for the arguments, and you end up validating messy input by hand.
• Reaching for an agent too early. If the model only needs to return data, structured outputs or a single tool call is simpler and cheaper than an agent loop.

Related

• Vercel AI SDK tool calling example: a lighter agent loop without a full framework.
• OpenAI function calling in JavaScript: the raw function-calling loop the frameworks wrap.

References

• LangChain JavaScript overview
• LangChain tools docs