Manage your Reedle reading library and extract content from the terminal or as an AI agent.
Tool preference: Use reedle CLI (Bash) for all operations — it's faster, token-efficient, and composable with grep, jq, and pipes. Use mcp__reedle__* MCP tools as fallback when the CLI is not installed or when complex multi-step MCP tool chains are more appropriate.
Install:npm install -g reedle-cli
Auth: Generate a CLI token at app.reedle.iopho.com → Settings → Integrations → CLI Token, then:
reedle auth token <rdk_xxx> # Set PAT (get from web Settings → Integrations → CLI Token)
reedle auth whoami# Verify token + show status
reedle auth logout# Remove stored credentials
Token stored in ~/.config/reedle/credentials.json (or ~/Library/Application Support/reedle/ on macOS).
Override with env var: REEDLE_TOKEN=rdk_xxx reedle list
CRITICAL: Save vs Read Decision Rule
Before using any save/read tool, determine user intent:
User says…
Use
"save", "add", "keep", "bookmark"
reedle save / reedle_save*
"read", "summarize", "check", "fetch", "what does this say"
reedle read / reedle_read*
Unclear
Default to reedle_read* — never pollute library without clear intent
After reedle read: Offer "This article was interesting — want me to save it to your Reedle library?"
Why this matters: The user's Reedle library is their curated knowledge base. Saving content they didn't ask to save degrades it. When in doubt, extract first, ask to save after.
Save to Library
Saves content permanently to the user's Reedle library. Full LLM pipeline runs (fetch → parse → embed). Processing is async — use reedle show <id> or reedle_get_processing_status to check when ready.
reedle save <url> # Save a web article
reedle save <url> -t research -t ai # Save with tags
MCP tools (when CLI not available):
mcp__reedle__reedle_save — save web article URL to library
mcp__reedle__reedle_save_youtube — save YouTube video (transcript) to library; params: video_url_or_id, language_code?
mcp__reedle__reedle_save_bilibili — save Bilibili video (transcript) to library; params: video_url_or_id, cookie (required), page?
mcp__reedle__reedle_get_processing_status — check save pipeline status
Read / Extract (ephemeral, no library save)
Extracts content inline without saving. Credit costs: web=5cr, YouTube=10cr, Bilibili=15cr, +20cr if --llm.
reedle read <url> # Extract web article → stdout
reedle read <url> --llm # With LLM cleanup (better quality, +20cr)
reedle read <url> | head -100 # First 100 lines
reedle read <url> --json # JSON with metadata
reedle read <youtube-url> # Auto-detects YouTube URLs
reedle read <video-id> --youtube # Force YouTube mode
reedle read <video-id> --youtube --lang zh # Chinese transcript
reedle read <youtube-url> --save # Extract + also save to library
reedle read <bvid> --bilibili --cookie "..."# Bilibili transcript
reedle list # List recent articles
reedle list --tag research # Filter by tag
reedle list --list <list-id> # Filter by list
reedle list --starred # Starred only
reedle list --status ready # Filter by status
reedle list -n 100 # Increase limit (default: 50)
reedle show <id> # Show article metadata
reedle cat <id> # Output full article text to stdout (pipe-friendly)
reedle open <id> # Open in browser
Actions
reedle star <id> # Star an article
reedle unstar <id> # Unstar an article
reedle archive <id> # Archive an article
reedle unarchive <id> # Unarchive an article
reedle delete <id> --force # Delete (--force required to confirm)
reedle tag <id> <tag> # Add a tag
reedle untag <id> <tag> # Remove a tag
reedle tags # List all tags with counts
reedle lists # List all reading lists
Browse TUI
Full-screen 3-pane terminal interface. No external tools required — markdown rendering is bundled with reedle-cli.
reedle browse # All articles
reedle browse --starred # Starred only
reedle browse -t research # Filter by tag
Layout: sidebar (views/tags/lists) | article list | preview pane, with fullscreen reader on Enter.
Key map:
Key
Action
j/k
Navigate in focused pane
Tab
Cycle focus: sidebar → list → preview
1/2/3
Jump to sidebar / list / preview
Enter
Open fullscreen reader
/
Live search (title, site, excerpt, tags)
s
Toggle star
a
Toggle archive
dd
Delete (press twice to confirm, Esc cancels)
t
Add tag (input prompt)
o
Open in browser
r
Refresh from API
b / Esc / q
Exit reader back to list
q
Quit TUI (from list view)
Agentic Patterns
Research without polluting library (use read_* tools)
# Extract and analyze without saving
reedle read <url> | wc -w # Word count before deciding to save# MCP pattern for research:# 1. mcp__reedle__reedle_read (url) — get content# 2. Analyze content# 3. Ask user: "Want me to save this to your library?"# 4. If yes: mcp__reedle__reedle_save (url)
Save and tag batch URLs
for url in https://arxiv.org/... https://papers.with...
do
reedle save "$url" -t papers -t 2025
done
YouTube workflow
# Just read the transcript:
reedle read https://youtube.com/watch?v=xxx --youtube
# Read then save:
reedle read https://youtube.com/watch?v=xxx --youtube --save
# MCP: read first, offer to save# mcp__reedle__reedle_read_youtube (video_url_or_id, language_code?)# → returns transcript inline# → offer: "Want me to save this video to your library?"# mcp__reedle__reedle_save_youtube (video_url_or_id, language_code?)
Pipe article content to another tool
reedle cat <id> | wc -w # Word count
reedle cat <id> | glow # Render markdown in terminal
reedle read <url> | head -50 # Preview before deciding to save
reedle read <url> --json | jq '.content'# Extract content field
Export article list as JSON for processing
reedle list --tag research --json | jq '.[] | {id, title, url}'
MCP Fallback (when CLI not available)
Use mcp__reedle__* tools directly for operations the CLI doesn't yet cover:
