TL;DR: đź“ą Watch the Complete Video Tutorial đź“ş Title: Claude Code Web is INSANE ⏱️ Duration: 613 👤 Channel: Creator Magic…
đź“‹ Table of Contents
Jump to any section (22 sections available)
đź“ą Watch the Complete Video Tutorial
đź“ş Title: Claude Code Web is INSANE
⏱️ Duration: 613
👤 Channel: Creator Magic
🎯 Topic: Claude Code Web
đź’ˇ This comprehensive article is based on the tutorial above. Watch the video for visual demonstrations and detailed explanations.
In this comprehensive walkthrough, AI expert Mike demonstrates how to harness the power of Claude Code Web, Google’s new Gemini API file search, and Apify to build a fully functional YouTube channel chatbot—without writing a single line of code yourself. This guide captures every step, insight, tip, and troubleshooting moment from Mike’s real-time build, giving you an end-to-end blueprint for creating your own retrieval-augmented generation (RAG) tool using the latest AI advancements of the week.
Why This Week’s AI Tools Are Game-Changers
Mike opens by highlighting three major developments that make this project possible:
- Claude Code is now available on the web—no local setup required.
- Claude Pro subscribers receive $250 in credits valid for 10 days.
- Claude Max subscribers get $1,000 in credits—a massive incentive to experiment.
Simultaneously, Google has rolled out a groundbreaking feature in the Gemini API: native file search. This is retrieval-augmented generation (RAG) built directly into the API, eliminating the need for manual chunking, vector databases, or complex retrieval pipelines. Just upload files and ask questions—Google handles the rest.
Introducing the $1 Million Apify Challenge
While building the tool, Mike leverages Apify, a web scraping and automation platform he’s been using regularly. He notes that Apify has just launched a $1 Million Challenge prize pool for developers who create innovative data extractors or tools using their platform. The official YouTube crawler actor used in this project is part of Apify’s ecosystem, and creators are encouraged to participate using the provided link.
Project Goal: A YouTube Channel Chatbot with Full Transcript Search
The objective is clear and practical: build a tool that
- Prompts the user for a YouTube channel URL.
- Fetches the channel’s videos from newest to oldest.
- Uses Apify’s official YouTube crawler to extract video titles and subtitles (transcripts).
- Creates a text file for each video.
- Uploads these files to Gemini API’s new file search feature.
- Provides a chat interface to query all transcripts using natural language.
Step 1: Setting Up Claude Code Web
To begin, Mike navigates to Claude Code Web and connects his GitHub account. After authorization, he’s ready to start coding—but with a twist: Claude will do all the coding.
Creating the Repository
Mike creates a new GitHub repository named Gemini API rag. He emphasizes that empty repositories don’t work well with Claude Code Web. The system needs at least one file to detect context and begin work.
The Critical Workaround: Using CLAUDE.md
When the initial attempt stalls (Claude gets stuck in a cloning loop), Mike discovers a key insight: you must add a CLAUDE.md file containing your full prompt. He pastes his detailed instructions into this file and commits it. Once this file exists, Claude Code Web successfully detects the project and begins execution.
Step 2: Crafting the Perfect Prompt for Claude
Mike’s prompt is highly specific and includes all necessary technical directives:
“Build a tool for any YouTube channel. Ask me for the channel URL and store videos from newest to older. Use Apify to grab video title and subtitles. Create a file for each video and add it to the Gemini API file search (new this week). When done, give me a chat box to chat with those video transcripts.”
He also includes critical implementation notes:
- Store environment variables for Gemini API key and Apify API token.
- Use the official Apify YouTube crawler actor.
- Reference Google’s new file search documentation and Apify docs.
Step 3: Watching Claude Code Web in Action
Once the CLAUDE.md file is committed, Claude Code Web springs to life:
- It scans the repository directory.
- Fetches relevant documentation from the web (Gemini API, Apify, RAG best practices).
- Generates a detailed to-do list of implementation tasks.
Files Automatically Generated by Claude
Within minutes, Claude produces a complete project structure, including:
requirements.txt— Python dependencies.env.example— Template for API keysmain.py— Core YouTube scraping and RAG logicREADME.md— Project documentation- Supporting scripts for video processing and file upload
Step 4: Reviewing and Merging Claude’s Pull Request
Claude Code Web automatically creates a pull request on GitHub with all generated code. Mike reviews the changes, confirms the inclusion of placeholders for his API keys, and merges the PR. He then clones the repository locally to test the tool.
Step 5: Configuring the Environment
Mike follows the setup instructions in the README:
- Creates a Python virtual environment.
- Installs dependencies from
requirements.txt. - Copies
.env.exampleto.env. - Pastes his Gemini API key and Apify API token into the
.envfile.
Step 6: First Test Run – Processing Mike’s YouTube Channel
Mike runs the script and enters the URL of his secondary YouTube channel. He chooses to process the 5 latest videos. The tool immediately:
- Triggers the Apify YouTube crawler in his Apify dashboard.
- Successfully scrapes video metadata and subtitles.
- Begins uploading transcript files to Gemini’s file search.
Step 7: Encountering and Debugging an API Error
The first run fails during the file upload to Gemini API. Mike captures the terminal error and feeds it back to Claude Code Web via a new prompt:
“We got this far and it broke. [Pastes error output]”
Claude’s Debugging Process
Claude Code Web responds by:
- Identifying the issue: incorrect syntax for the Gemini file search API.
- Searching Google’s official documentation for the correct implementation.
- Performing a web search for “Gemini API Python file upload chat example”.
- Generating a fix and creating a new pull request with corrected code.
Step 8: Successful RAG Chat with Personal Channel Data
After merging the fix and re-running the script, Mike tests the chat interface by asking: “What did Mike say about Adobe?”
Impressive Results Validate Accuracy
The response is highly accurate and specific:
- Confirms Mike is an Adobe Certified Professional.
- References his talk at Adobe Max 2025.
- Mentions his attendance at the IBC conference earlier in the year.
- Highlights new Adobe features like Firefly Generate Soundtrack (released just two weeks prior).
- Discusses Adobe Photoshop AI integrations.
Mike confirms this is not hallucination—the model is correctly retrieving and synthesizing information from his actual video transcripts.
Step 9: Testing on Another Creator’s Channel (Matt Wolfe)
To prove the tool’s versatility, Mike points it at fellow AI YouTuber Matt Wolfe’s channel. The system:
- Scrapes Matt’s 5 latest videos via Apify.
- Uploads transcripts to Gemini file search.
- Enables immediate querying.
Query Result: Top AI Trends from Matt Wolfe
When asked, “Tell me the top five things Matt mentioned recently,” the chatbot returns:
- ChatGPT Atlas
- Google Veo 3.1
- OpenAI Sora 2 updates
- Anthropic’s Claude Haiku 4.5
- Google’s Nano Banana
This demonstrates the tool’s ability to provide real-time, accurate summaries of any creator’s recent content.
The Power of Google’s Native File Search RAG
Mike emphasizes a key advantage: no more “spaghetti workflows”. Traditional RAG requires:
- Text chunking
- Embedding generation
- Vector database setup
- Custom retrieval logic
With Gemini’s new file search, Google handles all of this automatically. Developers simply upload files and query them—dramatically lowering the barrier to building intelligent applications.
Cost and Credit Management with Claude Code Web
Mike notes that the entire code generation process cost only $1 of his $250 Pro subscriber credit. This highlights the efficiency and affordability of using Claude Code Web for rapid prototyping, especially within the 10-day credit window.
Troubleshooting Tips from Real Experience
Based on his live debugging session, Mike provides these critical tips:
| Issue | Solution |
|---|---|
| Claude Code Web loops indefinitely on empty repo | Create a CLAUDE.md file with your prompt and commit it first |
| Gemini file upload fails with API error | Feed the full error output back to Claude; it will research and fix the syntax |
| Apify crawler not starting | Ensure your Apify API token is correctly set in the .env file |
Required Tools and Resources
Mike uses the following stack—all available this week:
| Tool | Purpose | Link/Note |
|---|---|---|
| Claude Code Web | AI-powered code generation in browser | Requires Pro/Max subscription for credits |
| Gemini API (with File Search) | Native RAG via file uploads | New feature released this week |
| Apify | YouTube video scraping and transcript extraction | Uses official YouTube crawler actor |
| GitHub | Code repository and Claude integration | Claude GitHub app must be installed |
How to Replicate This Project Yourself
Follow this exact sequence:
- Sign up for Claude Pro or Max to access credits.
- Go to Claude Code Web and connect GitHub.
- Create a new GitHub repo (e.g.,
youtube-rag-chat). - Add a
CLAUDE.mdfile with your detailed prompt (include Apify + Gemini instructions). - Let Claude generate the code and merge its pull request.
- Set up local environment and add your API keys.
- Run the script, enter a YouTube URL, and specify video count.
- Chat with the transcripts using natural language.
Why This Workflow Changes Everything for Creators
This tool empowers content creators, researchers, and marketers to:
- Instantly analyze any YouTube channel’s content without watching hours of video.
- Track competitor messaging or industry trends in real time.
- Repurpose their own content by querying past videos for quotes, data, or ideas.
- Build custom knowledge bases from video libraries with zero infrastructure.
Join the Apify $1 Million Challenge
Mike encourages viewers to explore Apify’s ecosystem further. If you’re skilled at building web data extractors or automation tools, consider entering the $1 Million Challenge. The same YouTube crawler actor used in this demo is a great starting point for your submission.
Final Thoughts from Mike
Mike concludes by expressing his genuine excitement for these new tools. He emphasizes that you no longer need to understand the inner workings of RAG or API integrations—just describe what you want, and AI like Claude Code Web will build it for you using the latest documentation.
He invites the community to:
- Build their own versions using the new Gemini file search.
- Share their creations in the comments.
- Join his free, open community (linked below).
Key Takeaways
- Claude Code Web can generate full-stack applications from a single prompt.
- Gemini’s native file search eliminates traditional RAG complexity.
- Apify provides reliable, structured YouTube data extraction.
- Always seed your repo with a
CLAUDE.mdfile to avoid startup issues. - Use error outputs as prompts to let Claude debug its own code.
Call to Action
Ready to build your own YouTube RAG chatbot? Start by accessing Claude Code Web with your Pro or Max subscription, gather your Gemini API key and Apify token, and follow Mike’s prompt structure. And if you create something cool with the new Gemini file search API, share it with the community—Mike wants to see what you build!
