crewai-transcriptapi

TranscriptAPI: hosted YouTube transcript + video-discovery API for AI agents. CrewAI tools edition. Also available as an n8n community node, an MCP server and agent skills.

Why this package

Give a CrewAI agent eyes and ears on YouTube without shelling out to yt-dlp, scraping HTML, or juggling the Google YouTube Data API and its quota. TranscriptAPI is a hosted backend built for this: transcript extraction, video search, and video metadata, wrapped here as three CrewAI tools that drop straight into an Agent's tools list.

This package is deliberately curated, not a full mirror of the API: three tools cover the core research loop (find a video, learn about it, read what was said in it). See The full TranscriptAPI surface below for the rest (channel browsing, playlists, community posts) via the REST API or MCP.

Installation

pip install crewai-transcriptapi

Credentials

You need a TranscriptAPI key (starts with sk_):

1. Create an account at transcriptapi.com: 100 free credits, no card (one-time); paid plans from $5/mo (1,000 credits).
2. Create an API key on the dashboard.
3. Export it:

export TRANSCRIPTAPI_API_KEY="sk_..."

Tools

Tool	What it does	Cost
TranscriptAPITool	Fetch the transcript of a YouTube video (full URL, youtu.be, embed, Shorts, or bare ID) as plain text or timestamped JSON, with metadata	1 credit/call
TranscriptAPISearchTool	Search YouTube for videos, channels, playlists, or movies, with sort/upload-date/duration/feature filters (first page only; this tool does not expose a continuation token for further pages)	1 credit/page
TranscriptAPIVideoMetadataTool	Pull a video's view/like counts, publish date, description, links, uploading channel, and optional details (duration, category, tags) or related videos, without pulling the transcript	1 credit/call

When to use what

Job	Tool
Need the spoken content of a video: summarize, quote, translate, fact-check	TranscriptAPITool
Need to discover videos, channels, playlists, or movies about a topic before pulling anything else	TranscriptAPISearchTool
Need facts about a video without pulling its transcript: view/like counts, publish date, description, duration, tags, related videos	TranscriptAPIVideoMetadataTool
Building a research pipeline: find candidates, then decide which ones are worth a full transcript	TranscriptAPISearchTool -> TranscriptAPIVideoMetadataTool -> TranscriptAPITool
Need a channel's profile, uploads, playlists, posts, or sections	Step up to the REST API or MCP server; not exposed by this package

Usage

Basic setup: search, then transcribe.

from crewai import Agent
from crewai_transcriptapi import TranscriptAPITool, TranscriptAPISearchTool

researcher = Agent(
    role="Video researcher",
    goal="Find and summarize YouTube content on a topic",
    backstory="Researches spoken video content via transcripts.",
    tools=[TranscriptAPISearchTool(), TranscriptAPITool()],
)

The agent can then search for videos on a topic and fetch each transcript for summarizing, quoting, translating or analysis.

Research crew example: search, screen, then transcribe

Adding TranscriptAPIVideoMetadataTool lets the agent screen search results (view count, publish date) before spending a credit on a transcript it doesn't need:

from crewai import Agent, Task, Crew
from crewai_transcriptapi import (
    TranscriptAPITool,
    TranscriptAPISearchTool,
    TranscriptAPIVideoMetadataTool,
)

researcher = Agent(
    role="Video researcher",
    goal="Find the most relevant, well-established videos on a topic and summarize them",
    backstory=(
        "Searches YouTube for candidates, checks each one's metadata to judge "
        "relevance and credibility before spending a transcript credit, then "
        "reads the transcripts of the ones worth reading."
    ),
    tools=[
        TranscriptAPISearchTool(),
        TranscriptAPIVideoMetadataTool(),
        TranscriptAPITool(),
    ],
)

task = Task(
    description=(
        "Find 3 well-established videos about {topic}. For each, check the "
        "view count and publish date before deciding it's worth summarizing, "
        "then fetch the transcript and write a 5-bullet summary."
    ),
    expected_output="A markdown report with one section per video.",
    agent=researcher,
)

Crew(agents=[researcher], tasks=[task]).kickoff(inputs={"topic": "transformer architectures"})

Calling a tool directly

Every tool returns a JSON string, so you can call it outside of a Crew for scripting or testing:

import json
from crewai_transcriptapi import TranscriptAPIVideoMetadataTool

tool = TranscriptAPIVideoMetadataTool()
result = json.loads(tool.run(video_url="UF8uR6Z6KLc", include="details,related"))
if result["success"]:
    print(result["data"]["title"], result["data"]["viewCountText"])

Use cases

• Check a video's facts without pulling its transcript. TranscriptAPIVideoMetadataTool returns view/like counts, publish date, description, and related videos.
• Beyond this package. TranscriptAPISearchTool (search_type="playlist") finds a playlist on a topic; listing that playlist's videos or browsing the channel that made it are REST/MCP operations (playlist/videos, channel/videos), not part of this package's 3 tools.

Tool reference

TranscriptAPITool

Argument	Type	Default	Description
video_url	str	required	Full YouTube URL (watch, youtu.be, embed, Shorts) or bare 11-character video ID
output_format	str	"text"	"text" for plain transcript, "json" for timestamped segments
language	str	none	Optional preferred transcript language code, for example "es"

TranscriptAPISearchTool

Argument	Type	Default	Description
query	str	required	Search query, 1 to 200 characters
search_type	str	"video"	"video", "channel", "playlist", or "movie"
sort	str	none	"relevance" or "views"
upload_date	str	none	"hour", "today", "week", "month", or "year" (videos only)
duration	str	none	"short" (<4m), "medium" (4-20m), or "long" (>20m) (videos only)
features	str	none	Comma-separated filters, e.g. "hd,subtitles,cc,live,4k,hdr,360,creative_commons"

TranscriptAPIVideoMetadataTool

Argument	Type	Default	Description
video_url	str	required	Full YouTube URL (watch, youtu.be, embed, Shorts) or bare 11-character video ID
include	str	none	Comma-separated extras: "details" and/or "related"

Costs

A successful call costs 1 credit: TranscriptAPITool per transcript, TranscriptAPISearchTool per call (first page only), TranscriptAPIVideoMetadataTool per lookup (include does not change the price). Failed calls and rate-limited (429) calls cost 0.

Response envelope

All three tools return a stable JSON string and never raise:

{"success": true, "data": {"transcript": "...", "metadata": {"title": "..."}}}

{"success": false, "error": {"code": "out_of_credits", "message": "The account is out of credits. See https://transcriptapi.com/billing."}}

Error codes: missing_api_key, invalid_api_key, out_of_credits, not_found, rate_limited, network, bad_response, http_<status>. Failed calls are never charged.

The full TranscriptAPI surface

This package ships 3 curated tools for the core research loop. TranscriptAPI itself covers more ground: channel profiles, uploads, Shorts, live streams, playlists, community posts, and curated channel sections. That breadth is available today via the REST API or the MCP server (12 tools), and via the n8n community node (11 operations). If your CrewAI agents need channel or playlist browsing, call the REST API directly with a custom tool, or open an issue.

Resources

• TranscriptAPI documentation
• REST API reference
• MCP reference
• Pricing
• Family: n8n node · MCP server · agent skills

Disclosure

TranscriptAPI is an independent product and is not affiliated with or endorsed by YouTube or Google. Use of these tools is subject to the TranscriptAPI terms.

License

MIT