DEV Community: MK

I turned a YouTube trading video into a backtested strategy — with Claude + MCP

MK — Wed, 10 Jun 2026 19:38:23 +0000

Most trading "strategies" on YouTube are locked inside long videos — often in another language, usually light on specifics, and frequently a runway to a paywall pitch. I wanted to see if I could pull the actual idea out of those videos and test it honestly, without watching hours of footage.

Here's a workflow I ran end to end, using Scribefy — MCP server for YouTube transcripts — as the front door.

1. Extract — turn video into something a model can reason over

I gave Claude a link to a trading video and had it call Scribefy to pull the full transcript. Because Scribefy handles non-English content, I worked through a Bulgarian creator's channel entirely in English — the intro video, then a live session, then the "theory" video where the mechanics were actually explained. Hours of watching collapsed into a few minutes of extraction.

2. Reverse-engineer — separate the mechanism from the narrative

With the transcripts in hand, Claude and I distilled the method to its codifiable core: a "Power-of-Three (PO3) dealing range" — a formula that partitions price into fixed levels. There's a lot of narrative wrapped around it in the videos; having the raw transcript let us isolate the part that's actually testable and set the rest aside.

3. Test — let the backtester decide, not the hype

This is the part that matters. We translated the logic into a Pine Script strategy for TradingView, hit the free-tier data limits, and moved to a MetaTrader 5 Expert Advisor (MQL5) for unlimited historical backtesting.

The non-negotiable: every version shipped with a built-in control — a deliberately misaligned grid — so we could tell a real edge from a good-looking fluke. (If a "strategy" can't beat its own scrambled control, it doesn't have an edge.)

The original idea came out marginal. But the process pointed us toward a session-based breakout variant that, on gold (XAUUSD), backtested at:

Profit factor 1.23
Sharpe 1.58
Max drawdown 6.5%
…and, crucially, it passed its falsification control.

The takeaway

Scribefy isn't "just a transcript tool" — it's the front door to a research pipeline. Extract the idea, codify it, and let the backtester (not the video's hype) decide whether it's real. The transcript is what makes the whole loop possible — especially when the best explanation is buried in a 40-minute video in a language you don't speak.

Try it: scribefy.app · setup guides at scribefy.app/guides

Not financial advice. These are historical backtest results on past data — not a promise of future performance, and nothing here is a recommendation to trade. The point of this post is the research **process, not the strategy.

YouTube transcripts in Cursor & Claude: a 2-minute MCP setup

MK — Tue, 09 Jun 2026 20:16:53 +0000

Your AI editor is great at code and useless at video. Ask Claude or Cursor about a 40-minute talk and it shrugs — it can't watch YouTube. So you do the dance: open the video, hit "show transcript," copy the wall of text, strip the timestamps, paste it in. Every. Single. Time.

MCP kills that loop — you give your agent a tool it can call, and it fetches the transcript itself, mid-conversation.

The setup: drop Scribefy into your MCP config (npx -y scribefy-mcp + SCRIBEFY_API_KEY), grab a key at scribefy.app, restart → tools appear. (links to the Cursor / Windsurf / Claude guides)

What your agent gets: extract_transcript, search_videos, get_video_metadata, get_related_videos.

3 ways I use it: summarize a talk with timestamps · lift commands out of a coding tutorial · search + compare across multiple videos in one prompt.

Why a tool beats a script: the model decides when to fetch, which video, which language — "I have my answer," not "I have a transcript."

https://scribefy.app/ - Free credits for signed in users

Here is something to talk about, Stop copy-pasting YouTube transcripts into Claude — wire it up with MCP

MK — Wed, 03 Jun 2026 09:01:13 +0000

Hey 👋 — sharing something I've been working on lately, in case it's useful to anyone doing similar stuff.

Quick context: I use Claude a lot for digging through YouTube — conference talks, tutorials, the occasional 2-hour podcast. And the workflow was always the same annoying dance: open the video, click "Show transcript," copy the wall of text, paste it into Claude, ask my question. Every. Single. Time. And you lose the timestamps, so when Claude says "they mention X around the middle," you can't easily jump back and check.

At some point it bugged me enough that I built a little thing to fix it. It's an MCP server — basically a small program that hands Claude (or Cursor / Windsurf) a set of YouTube tools so it can just… do the fetching itself. I called it Scribefy. (Heads up: it's my own project, so take the plug with the appropriate grain of salt — but the how-it-works bits apply to any YouTube MCP server.)

What it actually does

Once it's plugged in, the assistant gets four tools:

search_videos — search YouTube
get_video_metadata — title, channel, length, whether it even has captions
get_related_videos — the "up next" list
extract_transcript — the full transcript, with timestamps

So now I just paste a URL and say "summarize this with timestamps," or even "find me 3 videos on X and tell me where they disagree" — and it figures out the rest. No copy-paste.

Who it's actually handy for

It started as a "scratch my own itch" thing, but the use cases turned out way broader than I expected. A few that keep coming up:

Students — drop in a 90-minute lecture and get the 10 things you actually need for the exam, with timestamps to the parts worth rewatching. Beats scrubbing the progress bar hunting for that one slide you half-remember.
Researchers — treat a stack of talks/interviews like a mini literature review: "what do these five videos agree and disagree on?" — and get quotes you can cite back to the exact second.
Content creators — see what's already been said on a topic before you film, or turn one of your own videos into show notes, a blog post, and a thread in one pass. (The repurpose-one-video-into-five-formats move.)
Beginner traders — the forex/trading corner of YouTube is endless hour-long analysis videos. Instead of rewatching, you can pull out the actual setup, rules, and levels a video lays out and ask follow-up questions. (Not financial advice, obviously — just a much faster way to digest the educational stuff.)

Different people, same core trick: let the AI read the video so you don't have to sit through all of it.

If you want to poke at it

It's an npx package, same config in Claude Desktop, Cursor, and Windsurf:

{ "mcpServers": { "scribefy": { "command": "npx", "args": ["-y", "scribefy-mcp"], "env": { "SCRIBEFY_API_KEY": "sk_live_…" } } } }

(Claude Desktop → Settings → Developer → Edit Config. Cursor → ~/.cursor/mcp.json. Windsurf → ~/.codeium/windsurf/mcp_config.json, then Refresh.) Restart and the tools show up. There's a key + a couple of free credits to mess around with.

The part I actually think is neat

It's not really about one transcript. It's that the assistant can research a whole topic on its own — search for candidates, glance at the metadata to throw out the 3-hour ones, then only pull transcripts for the handful worth reading. Search and metadata are free, so poking around a topic stays cheap and you only "spend" on the videos you actually use.

And the timestamps matter more than I expected — Claude can go "at 12:40 they argue…" and I can click straight there to check, instead of trusting a vibe-summary of a 40-minute video.

The honest bits (because I hate posts that skip these)

It's YouTube-only, and the video needs captions (auto-generated counts). It does not do its own speech-to-text, so a caption-less video is a no-go.
There are free open-source YouTube MCP servers too. If you're cost-sensitive and don't mind the occasional breakage when YouTube changes something under the hood, those are a totally legit route. Mine's a hosted/paid thing that trades money for reliability (runs through a residential proxy so it doesn't get bot-blocked) plus the extra search/metadata tools.

So — not magic, not for everyone. But if YouTube is part of how you learn or work, having the AI do the watching-and-pulling has been a genuinely nice upgrade.

Where I'm at

It's pretty fresh — just got it listed in the official MCP registry and I'm slowly telling people. Honestly I'm still figuring out what folks actually want it to do, so if you try it (or you've built something similar), I'd love to hear what tools you'd want your agent to have for YouTube.

Repo: https://github.com/MKirovBG/scribefy-mcp
Setup notes for each assistant (incl. a ChatGPT route): https://scribefy.app/guides/youtube-transcripts-for-ai-assistants

Anyway — that's the thing. Back to building. 🛠️