content-bridge

Content Bridge - General-purpose content ingestion routing layer

Intelligent routing layer that automatically invokes the corresponding content ingestion skill based on the URL type, and returns standardized content.

Positioning and Responsibilities

Content Bridge is general-purpose content ingestion infrastructure, and is not bound to any downstream application.

Core responsibility: identify the content source type → route to the corresponding ingestion skill → return standardized content
Reusability: any skill (deep-learning, knowledge-sync, research, etc.) can depend on it
Not responsible for: it does not directly implement the scraping logic for specific platforms, and does not handle downstream application logic (such as NotebookLM upload)
Clear boundaries: focuses on the "content acquisition" layer; how downstream uses it is decided by the caller

Architecture Description

Content Bridge is arouting layer, not an implementation layer:

调用方 (deep-learning, etc.)
    ↓
content-bridge (路由判断)
    ↓
独立摄取 skills (weixin-extractor, bilibili-subtitle, etc.)
    ↓
标准化输出 (Markdown + metadata)

Dependent Ingestion Skills

Content Source	Ingestion Skill	Status
WeChat Official Account	`weixin-extractor`	✅ Created
Bilibili Video	`bilibili-subtitle`	✅ Exists
YouTube	`youtube-transcript`	✅ Created
General Web Page	`web-reader`	✅ Created
PDF/DOCX	`document-parser`	✅ Created

Supported Content Sources

1. WeChat Official Account Articles

Automatically scrape WeChat Official Account article content via the MCP server (bypassing anti-scraping)

2. Any Web Page Link

Supports any publicly accessible web page (news, blogs, documents, etc.)

3. YouTube Videos

Automatically extract subtitles and metadata from YouTube videos

4. Office Documents

Word (DOCX) - Preserves tables and formatting
PowerPoint (PPTX) - Extracts slides and notes
Excel (XLSX) - Tabular data

5. E-books and Documents

PDF - Full-text extraction
EPUB - Full-text extraction of e-books
Markdown (.md) - Native support

6. Images and Scanned Documents

Images (JPEG, PNG, GIF, WebP) - OCR text recognition
Scanned PDF documents - OCR text extraction

7. Audio Files

Audio (WAV, MP3) - Speech-to-text

8. Structured Data

CSV - Comma-separated data
JSON - JSON data
XML - XML documents

9. Archives

ZIP - Automatically decompresses and processes all supported files

10. Plain Text

Text content entered or pasted directly

11. Search Keywords

Searches keywords via Web Search and aggregates information from multiple sources

12. Bilibili Videos

Extracts subtitles or transcribes audio via the bilibili-subtitle tool to generate a text transcript

Prerequisites

1. Install wexin-read-mcp

The MCP server is already installed at:~/.claude/skills/anything-to-notebooklm/wexin-read-mcp/

Configure MCP(needs to be manually added to the Claude configuration file):

macOS: Edit ~/.claude/config.json

{
  "primaryApiKey": "any",
  "mcpServers": {
    "weixin-reader": {
      "command": "python",
      "args": [
        "/Users/joe/.claude/skills/anything-to-notebooklm/wexin-read-mcp/src/server.py"
      ]
    }
  }
}

After configuration, Claude Code needs to be restarted.

2. NotebookLM Authentication

Before first use, delegate to the notebooklm Skill to execute the authentication process:

Run authentication check (auth check)
If not logged in or expired, execute login (login)
Return authentication status to this Skill, and continue upload/generation only after confirmation

Trigger Methods

WeChat Official Account Article

/anything-to-notebooklm [微信文章链接]
"Upload this WeChat article to NotebookLM"
"Generate a podcast from this WeChat article"

Web Link

"Make this webpage into a podcast [URL]"
"Make this article into a PPT for me [URL]"
"Help me analyze this webpage [URL]"

YouTube Video

"Make this YouTube video into a podcast [YouTube URL]"
"Generate a mind map for this video [YouTube URL]"

Bilibili Video

"Make this Bilibili video into a podcast [Bilibili URL]"
"Generate a PPT for this Bilibili video [BV number]"

Local File

"Upload this PDF to NotebookLM /path/to/file.pdf"
"Generate a PPT from this Markdown file /path/to/file.md"
"Generate a podcast from this EPUB ebook /path/to/book.epub"
"Make this Word document into a mind map /path/to/doc.docx"
"Generate a Quiz from this PowerPoint /path/to/slides.pptx"
"Make this scanned PDF into a report /path/to/scan.pdf" (automatic OCR)

Search Keywords

"Search for 'AI development trends' and generate a report"
"Search for materials about 'quantum computing' and make a podcast"

Mixed Use

"Upload this article, this video, and this PDF together, and generate a report"

Natural Language → Delegated Action Mapping

What the user says	Recognized intent	Delegated action (executed by the `notebooklm` Skill)
"Generate a podcast" / "make it into audio" / "convert to speech"	audio	generate-audio
"Make a PPT" / "generate slides" / "make a presentation"	slide-deck	generate-slide-deck
"Draw a mind map" / "generate a mind map" / "make a map"	mind-map	generate-mind-map
"Generate a Quiz" / "create questions" / "make a quiz"	quiz	generate-quiz
"Make a video" / "generate a video"	video	generate-video
"Generate a report" / "write a summary" / "organize into a document"	report	generate-report
"Make an infographic" / "visualize"	infographic	generate-infographic
"Generate a data table" / "make a table"	data-table	generate-data-table
"Make flashcards" / "generate memory cards"	flashcards	generate-flashcards

If there is no explicit instruction, by default only upload without generating any content, and wait for the user's follow-up instructions.

Progress Updates

Provide clear status updates during execution:

⏳ 识别内容源类型...
✅ 检测到微信公众号文章
⏳ 抓取文章内容...
✅ 文章抓取成功：《深度学习的未来趋势》
⏳ 保存为临时文件...
✅ 已保存：/tmp/weixin_深度学习_1706234567.txt
⏳ 创建 NotebookLM 笔记本...
✅ 笔记本已创建：ID abc-123-def
⏳ 上传文件到笔记本...
✅ 上传完成，等待处理...
✅ 处理完成！
⏳ 生成播客...
✅ 播客生成完成！
⏳ 下载文件...
✅ Done!

Output Summary

After the task is completed, output in the following format:

Upload only (no generation)：

✅ 内容已上传到 NotebookLM！

📓 笔记本：[标题]
   ID: abc-123-def

📄 来源：filename.txt
   ID: source-xyz-789

💡 后续操作：
   - 默认继续使用本 Skill：直接说"基于这个 notebook 生成播客/PPT/Quiz"
   - 高级模式：切换 `notebooklm` Skill，对该 notebook 进行精细控制

Upload and generate：

✅ [内容类型]已转换为[输出格式]！

📄 来源：[标题/文件名]
👤 作者：[作者]（如有）
📅 日期：[日期]（如有）

🎙️ [输出格式]已生成：
📁 文件：/tmp/output_file.mp3
⏱️ 时长/页数：约 X 分钟/页
📊 大小：X MB

Workflow

Step 1: Identify the content source type

Claude automatically identifies the input type:

Input feature	Identified as	Processing method
`https://mp.weixin.qq.com/s/`	WeChat Official Account	MCP tool scraping
`https://youtube.com/...` or `https://youtu.be/...`	YouTube	Pass directly to NotebookLM
`https://bilibili.com/...` or `BV1xxx`	Bilibili video	bilibili-subtitle extract → TXT
`https://` or `http://`	Web page	Directly pass to NotebookLM
`/path/to/file.pdf`	PDF file	markitdown convert to Markdown → TXT
`/path/to/file.epub`	EPUB ebook	markitdown convert to Markdown → TXT
`/path/to/file.docx`	Word document	markitdown convert to Markdown → TXT
`/path/to/file.pptx`	PowerPoint	markitdown convert to Markdown → TXT
`/path/to/file.xlsx`	Excel	markitdown convert to Markdown → TXT
`/path/to/file.md`	Markdown	Direct upload
`/path/to/image.jpg`	Image (OCR)	markitdown OCR → TXT
`/path/to/audio.mp3`	Audio	markitdown transcribe → TXT
`/path/to/file.zip`	ZIP archive	Extract → markitdown batch conversion
Keyword (no URL, no path)	Search query	WebSearch → Summarize → TXT

Step 2: Get content

WeChat official account：

Use MCP tool read_weixin_article
Returns: title, author, publish_time, content
Save as TXT:/tmp/weixin_{title}_{timestamp}.txt

Web page/YouTube：

Keep the original URL as the source input
NotebookLM automatically extracts content

Bilibili video：

Use bilibili-subtitle to extract subtitles/transcript
Command:python -m bilibili_subtitle "URL" -o /tmp --skip-summary
Save as TXT:/tmp/{video_id}.transcript.md

Office document/ebook/PDF：

Use markitdown to convert to Markdown
Command:markitdown /path/to/file.docx -o /tmp/converted.md
Save as TXT:/tmp/{filename}_converted_{timestamp}.txt

Local Markdown：

Hand it to the delegation layer as a local file source for upload

Image (OCR)：

markitdown automatically performs OCR to recognize text
Extract EXIF metadata
Save as TXT

Audio file：

markitdown automatically transcribes speech to text
Extract audio metadata
Save as TXT

ZIP archive：

Automatically extract to a temporary directory
Traverse all supported files
Batch convert using markitdown
Merge into a single TXT or multiple Sources

Search keyword：

Use the WebSearch tool to search the keyword
Summarize the first 3-5 results
Save as TXT:/tmp/search_{keyword}_{timestamp}.txt

Step 3: Upload to NotebookLM

Delegate notebooklm Skill execution:

Create notebook (returns notebook_id)
Add source (file or URL)
Wait for the source to enter READY status
Pass notebook_id/source_id back to the subsequent steps of this Skill

Waiting for source READY is very important, otherwise subsequent generation may fail.

Step 5: Generate content based on intent (optional)

If the user has specified a processing intent, automatically delegate notebooklm Skill：

Intent	Delegated action	Artifact type
audio	generate-audio	mp3
slide-deck	generate-slide-deck	pdf
mind-map	generate-mind-map	json
quiz	generate-quiz	markdown/json/html
video	generate-video	mp4
report	generate-report	md
infographic	generate-infographic	png
flashcards	generate-flashcards	markdown/json/html

Generation process：

Delegate generation initiation (returns task/artifact ID)
Delegate waiting for completion
Delegate download to local
Inform the user of the final file path and size

Complete examples

Example 1: WeChat Official Account article → Podcast

User input：

把这篇文章生成播客 https://mp.weixin.qq.com/s/abc123xyz

Execution flow：

Recognized as a WeChat Official Account link
MCP tool fetches the article content
Create a TXT file
Upload to NotebookLM
Generate podcast (generate audio）
Download the podcast locally

Output：

✅ 微信文章已转换为播客！

📄 文章：深度学习的未来趋势
👤 作者：张三
📅 发布：2026-01-20

🎙️ 播客已生成：
📁 文件：/tmp/weixin_深度学习的未来趋势_podcast.mp3
⏱️ 时长：约 8 分钟
📊 大小：12.3 MB

Example 2: YouTube video → Mind map

User input：

这个视频帮我画个思维导图 https://www.youtube.com/watch?v=abc123

Execution flow：

Recognized as a YouTube link
Pass directly to NotebookLM (automatic subtitle extraction)
Generate mind map (generate mind-map）
Download the mind map

Output：

✅ YouTube 视频已转换为思维导图！

🎬 视频：Understanding Quantum Computing
⏱️ 时长：23 分钟

🗺️ 思维导图已生成：
📁 文件：/tmp/youtube_quantum_computing_mindmap.json
📊 节点数：45 个

Example 3: Search keyword → Report

User input：

搜索 'AI发展趋势 2026' 并生成报告

Execution flow：

Recognized as a search query
WebSearch searches the keyword
Summarize the top 5 results
Create a TXT file
Upload to NotebookLM
Generate report (generate report）

Output：

✅ 搜索结果已生成报告！

🔍 关键词：AI发展趋势 2026
📊 来源：5 篇文章

📄 报告已生成：
📁 文件：/tmp/search_AI发展趋势2026_report.md
📝 章节：7 个
📊 大小：15.2 KB

Example 4: Mixed multiple sources → PPT

User input：

把这篇文章、这个视频和这个PDF一起做成PPT：
- https://example.com/article
- https://youtube.com/watch?v=xyz
- /Users/joe/Documents/research.pdf

Execution flow：

Create a new Notebook
Add 3 Sources in sequence
Generate PPT based on all Sources

Output：

✅ 多源内容已整合为PPT！

📚 内容源：
  1. 网页文章：AI in 2026
  2. YouTube：Future of AI
  3. PDF：Research Notes (12 页)

📊 PPT 已生成：
📁 文件：/tmp/multi_source_slides.pdf
📄 页数：25 页
📦 大小：3.8 MB

Example 5: EPUB e-book → Podcast

User input：

把这本电子书做成播客 /Users/joe/Books/sapiens.epub

Execution flow：

Recognized as an EPUB file
markitdown converts to Markdown
Save as TXT
Upload to NotebookLM
Generate podcast

Output：

✅ EPUB 电子书已转换为播客！

📚 电子书：Sapiens: A Brief History of Humankind
📄 页数：约 450 页
📊 字数：约 15 万字

🎙️ 播客已生成：
📁 文件：/tmp/sapiens_podcast.mp3
⏱️ 时长：约 45 分钟（精华版）
📊 大小：48.2 MB

Example 6: Word document → Quiz

User input：

这个Markdown生成Quiz /Users/joe/notes/machine_learning.md

Execution flow：

Recognized as a local Markdown file
Upload directly to NotebookLM
Generate Quiz (generate quiz）

Output：

✅ Markdown 已转换为Quiz！

📄 文件：machine_learning.md
📊 大小：8.5 KB

📝 Quiz 已生成：
📁 文件：/tmp/machine_learning_quiz.md
❓ 题目：15 道（10选择 + 5简答）

Error handling

1. Failed to fetch WeChat article

Error: unable to fetch article content
Cause: article deleted, login required, network issue, anti-scraping block
Solution: check whether the link is correct, wait 2-3 seconds and retry
Fallback: manually copy the article content

2. NotebookLM authentication failed

Error：Auth/cookie error
Cause: authentication expired or not logged in
Solution: switch to the notebooklm Skill to run the authentication repair process (auth check/login)
Verify: authentication status is returned by the notebooklm Skill

3. File upload failed

Error：Invalid file or upload error
Cause: incorrect file path, permission issue, file too large
Solution: this Skill first checks the path/permissions/conversion result, then delegates to the notebooklm Skill to retry the upload
Fallback: try converting to a smaller file

4. Generation task failed

Error：Generation failed
Cause: Content is too short (<100 characters), too long (>500,000 characters), or the service is abnormal
Solution: By notebooklm Skill, perform task status diagnosis and retry
Alternative: Try other formats (e.g., a report instead of a podcast)

5. MCP tool not found

Error：Tool not found: read_weixin_article
Cause: The MCP server is not configured or not started
Solution: Check ~/.claude/config.json configuration, restart Claude Code
Verify：python ~/.agents/skills/anything-to-notebooklm/wexin-read-mcp/src/server.py

Advanced Features

1. Multi-intent processing

Users can specify multiple processing tasks at once:

这篇文章帮我生成播客和PPT https://mp.weixin.qq.com/s/abc123

The Skill will execute in sequence:

Generate podcast
Generate PPT

2. Custom Notebook

By default, a new Notebook is created for each article, but an existing Notebook can also be specified:

把这篇文章加到我的【AI研究】笔记本 https://mp.weixin.qq.com/s/abc123

The Skill will:

Search for a Notebook named "AI Research"
Add the article as a new Source
Generate content based on all Sources

3. Custom generation instructions

Add specific requirements for generation tasks:

这篇文章生成播客，要求：轻松幽默的风格，时长控制在5分钟

The Skill will pass the requirements as instructions to NotebookLM.

Notes

Rate limits：
- Interval between requests > 2 seconds to avoid being banned by WeChat
- NotebookLM generation tasks have concurrency limits (at most 3 running simultaneously)
Content length：
- WeChat articles are usually 1,000-5,000 characters, suitable for generating podcasts (3-8 minutes)
- Long articles over 10,000 characters may require longer generation time
- Short articles under 500 characters may produce poor results
Copyright compliance：
- For personal study and research only
- Comply with WeChat Official Account copyright regulations
- Generated content must not be used for commercial purposes
Generation time：
- Podcast: 2-5 minutes
- Video: 3-8 minutes
- PPT: 1-3 minutes
- Mind map: 1-2 minutes
- Quiz/flashcards: 1-2 minutes
File cleanup：
- TXT source files are saved in /tmp/, automatically cleaned up after a system restart
- Generated files (MP3/PDF/MD, etc.) are saved by default in /tmp/
- A custom save path can be specified

Related Skills

notebooklm - NotebookLM execution layer (advanced users can use directly)
notebooklm-deep-analyzer - In-depth analysis of NotebookLM content
markitdown - Convert documents in other formats

Default strategy：

By default, externally use anything-to-notebooklm(unified entry point)
For advanced scenarios, use notebooklmdirectly (fine-grained control)

Configure MCP (Important)

⚠️ Must be configured before first use

Edit ~/.claude/config.json：

{
  "primaryApiKey": "any",
  "mcpServers": {
    "weixin-reader": {
      "command": "python",
      "args": [
        "/Users/joe/.claude/skills/anything-to-notebooklm/wexin-read-mcp/src/server.py"
      ]
    }
  }
}

Restart Claude Code after configuration!

Troubleshooting

# 测试 MCP 服务器
python ~/.agents/skills/anything-to-notebooklm/wexin-read-mcp/src/server.py

Faults on the NotebookLM side (authentication, upload, generation, download) are uniformly delegated to notebooklm Skill to handle; a complete CLI troubleshooting manual is not maintained in this Skill.

Typical Use Cases

Scenario 1: Quick learning

我想学习这篇文章，帮我生成播客，上下班路上听
链接：https://mp.weixin.qq.com/s/abc123

→ Generate an 8-minute podcast to listen to during your commute

Scenario 2: Share with the team

这篇文章不错，做成PPT分享给团队
https://mp.weixin.qq.com/s/abc123

→ Generate a 15-page PPT, ready for team sharing

Scenario 3: Review and reinforce

这篇技术文章帮我出题，想测试一下掌握程度
https://mp.weixin.qq.com/s/abc123

→ Generate 10 multiple-choice questions + 5 short-answer questions

Scenario 4: Visual understanding

这篇文章概念比较多，画个思维导图帮我理清结构
https://mp.weixin.qq.com/s/abc123

→ Generate a mind map, clear at a glance

Example 7: Bilibili video → Podcast

User input：

把这个B站视频做成播客 https://www.bilibili.com/video/BV1234567890/

Execution flow：

Identified as a Bilibili link
bilibili-subtitle extracts subtitles/transcript
Save as TXT
Upload to NotebookLM
Generate podcast

Output：

✅ B站视频已转换为播客！

🎬 视频：深度学习入门教程
⏱️ 时长：45 分钟

🎙️ 播客已生成：
📁 文件：/tmp/bilibili_BV1234567890_podcast.mp3
⏱️ 时长：约 12 分钟
📊 大小：18.5 MB

Skill creation time：2026-01-25 Last updated：2026-02-25 Version：v1.2.0

About this skill

Positioning and Responsibilities

Architecture Description

Dependent Ingestion Skills

Supported Content Sources

1. WeChat Official Account Articles

2. Any Web Page Link

3. YouTube Videos

4. Office Documents

5. E-books and Documents

6. Images and Scanned Documents

7. Audio Files

8. Structured Data

9. Archives

10. Plain Text

11. Search Keywords

12. Bilibili Videos

Prerequisites

1. Install wexin-read-mcp

2. NotebookLM Authentication

Trigger Methods

WeChat Official Account Article

Web Link

YouTube Video

Bilibili Video

Local File

Search Keywords

Mixed Use

Natural Language → Delegated Action Mapping

Progress Updates

Output Summary

Workflow

Step 1: Identify the content source type

Step 2: Get content

Step 3: Upload to NotebookLM

Step 5: Generate content based on intent (optional)

Complete examples

Example 1: WeChat Official Account article → Podcast

Example 2: YouTube video → Mind map

Example 3: Search keyword → Report

Example 4: Mixed multiple sources → PPT

Example 5: EPUB e-book → Podcast

Example 6: Word document → Quiz

Error handling

1. Failed to fetch WeChat article

2. NotebookLM authentication failed

3. File upload failed

4. Generation task failed

5. MCP tool not found

Advanced Features

1. Multi-intent processing

2. Custom Notebook

3. Custom generation instructions

Notes

Related Skills

Configure MCP (Important)

Troubleshooting

Typical Use Cases

Scenario 1: Quick learning

Scenario 2: Share with the team

Scenario 3: Review and reinforce

Scenario 4: Visual understanding

Example 7: Bilibili video → Podcast