langchain-easytranscriber

LangChain integration for EasyTranscriber: load YouTube transcripts as Documents, and give agents tools to get a transcript, summarize a video, or search YouTube.

• EasyTranscriberLoader: YouTube URLs to Documents (transcript text plus metadata, optional AI summary)
• EasyTranscriberTranscriptTool: transcript of a YouTube video
• EasyTranscriberSummaryTool: Markdown AI summary of a YouTube video, in 65 languages
• EasyTranscriberSearchTool: YouTube video search, up to 10 results
• EasyTranscriberToolkit: the three tools with one API key

Transcripts come from the video's captions, or from speech-to-text when a video has no captions. Videos up to 4 hours are supported.

Install

pip install -U langchain-easytranscriber

Python 3.10+, langchain-core 1.x. Built on the easytranscriber-sdk client.

API key

Create a free account at easytranscriber.com and a key in Dashboard > API Keys. Signup includes 5 credits, and 1 credit is added each day while the balance is below 5. API access is included on every plan.

export EASYTRANSCRIBER_API_KEY="et_..."

Or pass api_key="et_..." to any class.

Credits

Plans and prices: pricing.

Document loader

from langchain_easytranscriber import EasyTranscriberLoader

loader = EasyTranscriberLoader(
    [
        "https://www.youtube.com/watch?v=dQw4w9WgXcQ",
        "https://youtu.be/9bZkp7q19f0",
    ],
    include_summary=True,  # adds metadata["summary"], 1 credit per video
    summary_language="English",  # omit to match each video's language
)

for doc in loader.lazy_load():
    print(doc.metadata["title"], doc.metadata["credits_used"])
    print(doc.page_content[:200])

Each Document has the transcript as page_content and this metadata:

Use continue_on_failure=True to skip videos that fail instead of raising. await loader.aload() and loader.alazy_load() use the async client.

Accepted URLs: youtube.com/watch?v=, youtu.be/, /shorts/ and /embed/ links. Bare video IDs are not accepted.

Tools

from langchain_easytranscriber import EasyTranscriberTranscriptTool

tool = EasyTranscriberTranscriptTool()
print(tool.invoke({"url": "https://www.youtube.com/watch?v=dQw4w9WgXcQ"}))

Invoked with a ToolCall, each tool returns a ToolMessage whose artifact holds the API response, including credits_used:

message = tool.invoke(
    {
        "args": {"url": "https://www.youtube.com/watch?v=dQw4w9WgXcQ"},
        "id": "1",
        "name": tool.name,
        "type": "tool_call",
    }
)
message.artifact["cached"], message.artifact["credits_used"]

In an agent

# pip install -qU "langchain[anthropic]"
from langchain.agents import create_agent
from langchain_easytranscriber import EasyTranscriberToolkit

agent = create_agent(
    model="anthropic:claude-sonnet-5-5",
    tools=EasyTranscriberToolkit().get_tools(),
)
result = agent.invoke(
    {
        "messages": [
            {"role": "user", "content": "Find a recent video on Python asyncio and summarize it."}
        ]
    }
)
print(result["messages"][-1].content)

API errors (for example insufficient_credits or unauthorized), timeouts and connection errors are raised as ToolException. Pass handle_tool_error=True to a tool to return the message to the agent instead.

Development

uv run --group test pytest                      # unit tests, mocked HTTP, no credits
uv run --group lint ruff check . && uv run --group lint ruff format --check .
uv run --group typing mypy
EASYTRANSCRIBER_API_KEY=et_... uv run --group test pytest tests/integration_tests  # live, spends credits

Unit tests include LangChain's standard ToolsUnitTests from langchain-tests.

Links

• REST API docs: https://www.easytranscriber.com/docs/api
• Python and JavaScript SDKs: https://www.easytranscriber.com/sdk
• MCP server: https://www.easytranscriber.com/docs/mcp

EasyTranscriber is not affiliated with YouTube or Google. MIT licensed.