langchain-string

String Web Access tools for LangChain.

Search the web and fetch any URL as clean, LLM-ready Markdown.

Installation

pip install -U langchain-string
# or
uv add langchain-string

Credentials

Get an API key at usestring.ai and set it:

export STRING_API_KEY=...

Or pass it per tool: StringSearch(api_key=SecretStr("...")).

Tools

Tool	What it does
StringSearch	Search Google, DuckDuckGo, Brave or Mojeek and return the results as JSON, with Google's knowledge panel, local listings and AI overviews when the page carried them.
StringFetch	Fetch any URL and return the page as Markdown.

Both implement _run and _arun, so they work in sync and async agents. API and network failures come back to the model as the tool's output, with the API's reason attached, rather than ending the run; pass handle_tool_error=False to raise them instead.

StringSearch

from langchain_string import StringSearch

search = StringSearch(max_results=5)
print(search.invoke({"query": "python httpx library"}))

StringFetch

from langchain_string import StringFetch

fetch = StringFetch(main_content_only=True)
print(fetch.invoke({"url": "https://example.com"}))

Pass execute_js=True when a JavaScript-rendered site comes back empty, and country_code="GB" to route the request through a specific country. When the site answers with an error status, such as a 404, the output says so above the page it returned.

Invoke with a ToolCall

Invoked with a model-generated ToolCall, either tool returns a ToolMessage:

fetch.invoke(
    {
        "args": {"url": "https://example.com"},
        "id": "1",
        "name": fetch.name,
        "type": "tool_call",
    }
)

Use within an agent

StringWebAccessToolkit returns both tools and passes its configuration to each.

# pip install -U "langchain[anthropic]" to run the agent and call the model
from langchain.agents import create_agent
from langchain_string import StringWebAccessToolkit

agent = create_agent(
    model="anthropic:claude-sonnet-5",
    tools=StringWebAccessToolkit(max_results=5).get_tools(),
    system_prompt=(
        "You research questions using the live web. Search first, then read the pages "
        "worth reading, and cite every URL you used."
    ),
)

result = agent.invoke(
    {
        "messages": [
            {
                "role": "user",
                "content": "Find the latest Python release and summarize what changed.",
            }
        ]
    }
)
print(result["messages"][-1].content)

Configuration

Shared by both tools:

Field	Default	Description
api_key	STRING_API_KEY	Your String API key, held as a SecretStr.
base_url	https://request.usestring.ai/v1	Point this elsewhere only for a different String deployment.
timeout	600.0	HTTP timeout in seconds. A browser-rendered fetch can take minutes.
http_client / http_async_client	—	Your own httpx clients, for proxies or custom transports.

StringSearch adds max_results, 1 to 50. Unset, a search returns one results page, about ten results. Google is paged until it has max_results, and each page is billed as one search.

StringFetch adds markdown_mode (full or readable), main_content_only and max_content_length.

Development

uv sync
make test                              # offline; network sockets are blocked
STRING_API_KEY=... make integration_test   # live, billed to that key
make lint

Releasing

Bump version in pyproject.toml on main. .github/workflows/publish.yml builds the wheel and sdist and publishes them to PyPI through trusted publishing, which needs a pending publisher configured on PyPI first.

Docs

• LangChain guide: https://portal.usestring.ai/docs/guides/langchain
• REST API: https://portal.usestring.ai/docs/api-reference/overview
• MCP server: https://portal.usestring.ai/docs/mcp/overview

License

MIT