CDP vs API Publishing: Which Platforms Welcome Automation in 2026

#automation #cdp #api #publishing

TL;DR: Map of every platform I tested for programmatic publishing in 2026: which give you a clean REST API, which need Chrome DevTools Protocol (CDP) automation, and which actively block both. Based on 50+ articles published over 60 days.

Three categories

Category	Platforms	Best approach
Clean REST API	dev.to, Substack (write API limited), Gumroad (limited), Mastodon, GitHub	API + auth header
CDP-friendly editors	Substack (TipTap), Gumroad UI, ASC web, WeChat 公众号 (cookie-bound)	Playwright connect_over_cdp
Blocked by anti-bot	Reddit, Indie Hackers, Twitter/X (Lexical), Hacker News, Zhihu, Xiaohongshu	Manual paste only

Why CDP, not Selenium/Puppeteer?

Selenium and headless Puppeteer get fingerprinted instantly. CDP (--remote-debugging-port=9222 --user-data-dir=...) attaches to a real Chrome you already opened with a real session.

Setup:

chrome.exe --remote-debugging-port=9222 ^
  --remote-allow-origins=* ^
  --user-data-dir="C:\Users\you\AppData\Local\Google\Chrome\User Data\AutomationProfile"

Then in Python:

from playwright.sync_api import sync_playwright
with sync_playwright() as p:
    browser = p.chromium.connect_over_cdp("http://127.0.0.1:9222")
    page = browser.contexts[0].pages[0]
    page.goto("https://your-target.com/editor")

The browser sees your real cookies, real fingerprint, real user-agent. No bot signal.

Editor difficulty rankings (1=easy, 5=blocked)

Editor	Used by	Difficulty	Notes
Plain `<textarea>`	dev.to drafts, basic CMS	1	`page.fill()` works
TipTap	Substack, Notion-likes	2	Click into ProseMirror, then Ctrl+V clipboard paste
Slate	Reddit (some), older Medium	3	Selection-based; needs careful focus management
Lexical (Meta's)	Reddit comments, IH posts, Twitter	5	Anti-bot detects synthetic events; only manual works
ProseMirror w/ Lexical input	Twitter post composer 2026	5	Same as above

What worked: Substack TipTap CDP flow

After 6 failed attempts on Substack (frontmatter leakage, body cut off, publish button not firing), this 5-step flow now publishes reliably:

Goto /publish/posts?type=newsletter → click 新規投稿 button
Fill title via page.fill('textarea[aria-label="title"]', title)
Click into editor (div.ProseMirror), set clipboard with body, dispatch Ctrl+V
Wait for "下書き保存しました" (autosave indicator)
Click 続ける → 今すぐ全員に送信 → confirm → URL changes to /p/<slug>

Total time per issue: ~45 seconds. 100% reliable across 4 issues.

What failed: Reddit Lexical

Tried 8 different approaches:

page.fill() on Lexical contenteditable → ignored
page.evaluate() to set innerHTML → reverted on focus
Clipboard paste via Ctrl+V → blocked by beforepaste listener
Synthetic InputEvent('beforeinput', {data:'...'}) → ignored
Direct DOM mutation → reverted by Lexical reconciler
Chrome accessibility API typing → flagged as bot
Per-character keyboard.press → rate-limited after 10 chars
CDP Input.dispatchKeyEvent → ignored

Reddit actively detects synthetic input on Lexical. Same for IH (Lexical) and Twitter post composer (Lexical-derived).

Conclusion: For Lexical platforms, paste-ready files + 30-second human paste is the only path.

What I built: paste-ready content pipeline

Since Lexical platforms are blocked, I built a unified pipeline:

CC writes paste-ready files with YAML frontmatter
Dashboard reads frontmatter, renders preview
Dashboard "Copy to clipboard" button copies body
User clicks dashboard "Open editor" → 30-second paste

This converts 4 manual platforms (Reddit/IH/Twitter/Zhihu) to:

4 × 30 sec = 2 minutes user time per content piece
vs 4 × 5 min manual writing each = 20 minutes

10× speedup on the human side, even though I can't fully automate.

The 2026 reality: hybrid is the answer

For indie hackers in 2026 publishing across 8+ channels:

Pure API: dev.to (4 hours of automated batch publishing)
CDP: Substack, Gumroad, ASC (all editor-driven)
Paste-ready: Reddit, IH, Twitter, Zhihu (anti-bot — never going to fully automate)

Stop trying to fully automate Lexical. Build a paste-ready pipeline instead.