> ## Documentation Index
> Fetch the complete documentation index at: https://docs.context.dev/llms.txt
> Use this file to discover all available pages before exploring further.

# LangChain

> Use Context.dev as native LangChain tools for live web data, documents, brand intelligence, monitors, and batches.

The [`langchain-context`](https://pypi.org/project/langchain-context/) package exposes Context.dev operations as LangChain `BaseTool` classes. Use individual tools directly or give an agent a focused toolkit for live web research, scraping, crawling, extraction, document parsing, brand intelligence, monitors, and batch jobs.

<Card title="View Context.dev for LangChain" icon="github" href="https://github.com/context-dot-dev/langchain-context">
  Inspect the package source, examples, tests, and releases.
</Card>

## Prerequisites

* Python 3.10 through 3.14
* A [Context.dev account](https://context.dev/signup) with an API key and credits for tool calls

## Install and configure

<Steps>
  <Step title="Install the package">
    ```bash theme={null}
    pip install -U langchain-context
    ```
  </Step>

  <Step title="Set your API key">
    Copy a secret key from the [Context.dev dashboard](https://context.dev/dashboard), then set it in your server environment:

    ```bash theme={null}
    export CONTEXT_API_KEY="ctxt_secret_your_key"
    ```

    Tools read `CONTEXT_API_KEY` automatically. You can also pass `api_key` to a tool or toolkit when your application already manages secrets securely.
  </Step>
</Steps>

<Warning>
  Keep `CONTEXT_API_KEY` in your server environment. Do not commit it or expose it in browser code.
</Warning>

## Use a tool directly

Every Context.dev operation supports LangChain's synchronous `invoke` interface:

```python theme={null}
from langchain_context import ContextScrape, ContextSearch

search = ContextSearch()
results = search.invoke(
    {
        "query": "latest official Stripe product announcements",
        "includeDomains": ["stripe.com"],
        "numResults": 10,
    }
)

scrape = ContextScrape()
page = scrape.invoke({"url": "https://stripe.com/newsroom"})
```

To read a YouTube video, pass its URL to `ContextScrape`. The returned Markdown includes the timestamped transcript when one is available.

## Give tools to an agent

`ContextToolkit` is safe by default: it includes read operations and removes browser actions from scraping tools. Pass its result to any LangChain agent that accepts a list of tools.

```python theme={null}
from langchain_context import ContextToolkit

context_tools = ContextToolkit(
    groups=("web", "brand"),
).get_tools()
```

Choose only the groups the agent needs:

| Group      | Includes                                                                                                                   |
| ---------- | -------------------------------------------------------------------------------------------------------------------------- |
| `web`      | Search, Markdown and HTML scraping, YouTube transcripts, crawling, extraction, screenshots, sitemaps, and document parsing |
| `brand`    | Brand profiles, style guides, fonts, and industry classification                                                           |
| `monitors` | Read existing monitors, runs, changes, and credit usage                                                                    |
| `batches`  | Read existing batch jobs and results                                                                                       |

Use a focused toolkit when you do not need every group:

```python theme={null}
from langchain_context import ContextBrandToolkit, ContextWebToolkit

web_tools = ContextWebToolkit().get_tools()
brand_tools = ContextBrandToolkit().get_tools()
```

## Enable side effects explicitly

Monitor and batch mutations are excluded unless you opt in. Browser actions are also removed from scraping tools by default.

```python theme={null}
from langchain_context import ContextToolkit

tools = ContextToolkit(
    include_write_tools=True,
    allow_browser_actions=True,
).get_tools()
```

Enable these options only for an agent that should create, update, run, cancel, or delete Context.dev resources or interact with third-party webpages.

## Use async tools

Every tool also supports LangChain's asynchronous `ainvoke` interface:

```python theme={null}
from langchain_context import ContextSearch

result = await ContextSearch().ainvoke(
    {"query": "latest Context.dev product announcements"}
)
```

## Troubleshooting

| Symptom                                     | Fix                                                                                       |
| ------------------------------------------- | ----------------------------------------------------------------------------------------- |
| `CONTEXT_API_KEY` is missing                | Set the environment variable before starting the application, or pass `api_key` securely. |
| Context.dev returns an authentication error | Replace the key with an active secret key from the Context.dev dashboard.                 |
| A mutating tool is not available            | Set `include_write_tools=True` only when the agent should change monitors or batches.     |
| Scraping does not expose browser actions    | Set `allow_browser_actions=True` only when interactive browsing is required.              |
| Calls fail for insufficient credits         | Review usage in the [Context.dev dashboard](https://context.dev/dashboard).               |

## Reference

<CardGroup cols={2}>
  <Card title="PyPI package" icon="python" href="https://pypi.org/project/langchain-context/">
    Install the current release and review its Python requirements.
  </Card>

  <Card title="Package source" icon="github" href="https://github.com/context-dot-dev/langchain-context">
    Read the complete tool catalog, examples, and test suite.
  </Card>

  <Card title="LangChain tools" icon="wrench" href="https://docs.langchain.com/oss/python/langchain/tools">
    Learn how tools are defined, invoked, and attached to agents.
  </Card>

  <Card title="Context.dev API reference" icon="code" href="/api-reference/web-scraping/search">
    Review the underlying API operations and request schemas.
  </Card>
</CardGroup>