Changelog
What shipped in the API, the MCP server, and the SDKs. For live uptime and response times, see status.
Follow along: RSS feed, or watch the widget's Changelog tab.
TranscriptFetch is in Claude's connector directory
You can now add TranscriptFetch straight from Claude's connector directory: search for TranscriptFetch, click Connect, and sign in. There's no server URL to paste. Once connected, Claude can get the transcript of a YouTube, TikTok or Instagram video, search for videos, and list channels, profiles and playlists, with AI transcription when a video has no captions. Every tool is read-only. It uses the same credits as the API, and a free account includes 50 credits a month.
Monitors: get every new video from a channel or profile
POST /api/v2/monitors watches a YouTube channel or a TikTok or Instagram profile and reports each new video as an event. The event is sent to your webhook, signed with HMAC-SHA256, and is always readable from the events endpoint. Add transcripts: true and each event carries the new videos' transcripts. No webhook? Every monitor returns has_new and last_event_id. Pass the last id back as ?since= and has_new tells you whether anything arrived after it. A check that finds nothing is free; one that finds new videos costs 1 credit, plus the usual transcript charge.
YouTube listing options: tabs, sorting and search filters
/channel now takes tab (videos, shorts, live, playlists or podcasts), sort (newest, popular or oldest), and query to search inside one channel. /search takes type (video, channel or playlist), upload_date, duration, sort, and captions: true for videos with subtitles. Playlists now page past YouTube's 200-video limit to their real end.
Two new MCP tools: search_channels and list_channel_playlists
The MCP server now has seven tools. An agent can find a creator's channel by name, list their playlists, and use the new YouTube listing options without leaving the conversation.
Transcription jobs report real progress
While an audio transcription job runs, its status includes a progress object. It shows the step it's on and the pieces done out of the total, with a percent that only moves on those real counts.
Free tier moves to 50 credits
New accounts now start with 50 free credits instead of 100, and accounts without a plan refresh to 50 credits each month. Everything else about the free tier is unchanged: no card required, 1 credit per successful transcript or listing page, 1 credit per minute of AI transcription, and failures never cost anything. Fifty credits is still enough to integrate and test every endpoint. When you need more, Basic is $5 for 1,000 credits a month, and unused plan credits roll over up to twice your plan.
YouTube search, playlists and channels are 3x faster
YouTube discovery is faster across the board. Keyword search, playlist listings and channel listings now return in well under a second at the median, down from about 2 seconds, and the slow tail is gone: 90th-percentile latency dropped from 3 to 5 seconds to around 1.2 seconds. Long playlists and channels benefit most. Paging deep into a playlist no longer slows down as you go, and a class of intermittent empty errors on listing calls is fixed. In benchmarks on real customer queries, playlist success went from 89% to 100%. Same request and response shapes, same 1 credit per listing page, nothing to change on your end.
AI transcription is now 1 credit per minute
AI audio transcription, the path used when a video or episode has no caption track, now bills 1 credit per started minute of audio, down from 2. A 10-minute video is 10 credits, an hour is 60, and the 4-hour maximum is 240. Nothing else moves: caption fetches stay 1 credit, listing pages stay 1 credit, and a job that fails still costs nothing. The new rate applies to every account from today, including the 100 free credits a month, and it changes what a plan buys: about 17 hours of audio for $5 on Basic, 167 hours for $15 on Pro, 833 hours for $45 on Mega, and 5,000 hours for $229 on Scale. If you transcribe a lot of audio, the same plan now goes twice as far. If you are on a paid plan and used AI transcription at the old rate since per-minute billing began on August 24, the difference has been credited back to your balance automatically. If you only ever want caption text, send mode: "captions" and a video without captions fails instead of being transcribed.
AI transcription just got seriously fast
We rebuilt the AI transcription pipeline for speed, and the difference is dramatic. Videos without captions used to mean a noticeable wait. Now a typical short-form video transcribes in a couple of seconds, about 3x faster than before, and the unpredictable slow requests are gone completely: in benchmarks on real TikTok, Shorts, and Reels content, the worst case dropped from over 20 seconds to under 4. Faster on long audio too. Same API, same pricing, no changes needed on your end. Your no-caption videos just come back at caption speed now.
More credits on every plan, and a new Scale tier
Basic now includes 1,000 credits a month at the same $5 (up from 500), Pro includes 10,000 at $15 (up from 5,000), Mega stays at $45 for 50,000, and a new Scale plan offers 300,000 credits for $229. Top-ups are also cheaper: instead of a flat $10 per 1,000, they now match your plan's own rate — $5 per 1,000 on Basic, $1.50 on Pro, $0.90 on Mega, $0.76 on Scale. Already subscribed? You get the new numbers: existing plans are topped up to the new credit balance automatically, at no extra charge. The 100 monthly free credits on every account are unchanged.
AI transcription moves to per-minute pricing
AI audio transcription (used when a video has no caption track) now bills 2 credits per minute of audio, matching the market rate for AI transcription APIs. For current subscribers the new rate takes effect at the start of their next billing period. Caption fetches stay 1 credit, failures stay free, and across the platform AI transcription typically accounts for about 20% of usage.
New MCP tool: get_credits
get_credits returns your remaining balance and is free to call, so an agent can check headroom before spending. The MCP server exposes every tool the REST API supports.
Typed tool outputs and read-only annotations
Every MCP tool now declares an output schema, so clients can type-check responses, plus MCP annotations marking all tools read-only and non-destructive. Added support for protocol revision 2025-11-25 and session-id headers for gateway compatibility.
GET /api/v1/me
A free endpoint that returns your user id and current credit balance. Useful for validating a key without spending a credit, which is how the n8n node tests credentials.
Official n8n community node
n8n-nodes-transcriptfetch is on npm with build provenance: fetch transcripts and search or enumerate videos from any n8n workflow, including as an AI Agent tool.
Monthly credit plans
Basic ($5 / 500 credits), Pro ($15 / 5,000) and Mega ($45 / 50,000), plus top-ups at $10 per 1,000 credits when you need more mid-month. One credit is one successful response; failures are never billed.
A page per error code
Every error the API can return now has its own page with the likely causes, how to fix it, and whether the request is safe to retry.
Transcripts beyond YouTube
TikTok and Instagram video URLs work on the same transcript endpoint as YouTube, with AI transcription when a video has no captions. Same request shape, same response shape.
Python and Node SDKs
Official clients for both languages, so you can install a package instead of hand-rolling HTTP calls and pagination.
MCP server
Connect TranscriptFetch to Claude, ChatGPT, Cursor, or any MCP client at https://transcriptfetch.com/mcp. OAuth sign-in creates an account with free credits, or pass an API key as a bearer token.
Built and maintained by one developer, shipping most days. Something broken or missing? Email me and you'll get a reply from the person who wrote the code.