reedle — Intelligent Reading Companion CLI

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: cd apps/reedle-cli && pnpm build && npm install -g . (monorepo) or npm install -g reedle

Auth: Generate a CLI token at app.reedle.iopho.com → Settings → Integrations → CLI Token, then:

reedle auth token rdk_xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx

Global Flags

Flag	Description
`--json`	Output as JSON (for scripting / jq)
`--no-color`	Disable color output
`-V, --version`	Show version

Auth

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 get <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

MCP tools (when CLI not available):

mcp__reedle__reedle_read            — extract web content; params: url, llm_cleanup?
mcp__reedle__reedle_read_youtube    — extract YouTube transcript; params: video_url_or_id, language_code?
mcp__reedle__reedle_read_bilibili   — extract Bilibili transcript; params: video_url_or_id, cookie (required), page?

Articles (Library Management)

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 get <id>                            # Get article metadata
reedle get <id> --content                  # Get full article text (markdown)
reedle open <id>                           # Open in browser

Search

reedle search "machine learning"           # Keyword search (title + excerpt)
reedle search "attention mechanism" -s     # Semantic (AI) search
reedle search "neural nets" -n 10          # Limit results
reedle search "topic" --json | jq '.'     # JSON output for scripting

Tags & Lists

reedle tags                                # List all tags with counts
reedle lists                               # List all reading lists

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 get <id> --content | wc -w         # Word count
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:

mcp__reedle__reedle_list_highlights — article highlights
mcp__reedle__reedle_list_comments — comments
mcp__reedle__reedle_find_similar_articles — similarity search
mcp__reedle__reedle_list_decks / reedle_list_cards_due — flashcards
mcp__reedle__reedle_get_credit_balance — credit balance

Credit Costs (Extraction Only)

Operation	Credits	Notes
`reedle_read` (web)	5 cr	Jina Reader fetch
`reedle_read_youtube`	10 cr	Transcript via reedle-server-py
`reedle_read_bilibili`	15 cr	Requires cookie
`+llm_cleanup` add-on	+20 cr	Gemini Flash content cleanup
Save operations	Dynamic	LLM cost deducted by server post-completion

1 credit = $0.001 USD. Check balance: mcp__reedle__reedle_get_credit_balance

Config

Item	Value
Token env var	`REEDLE_TOKEN`
Config dir	`~/.config/reedle/`
Credentials file	`~/.config/reedle/credentials.json`
API endpoint override	`REEDLE_API_ENDPOINT=<url>`
Web app	`https://app.reedle.iopho.com`
MCP endpoint	`https://fhgxapmrciwlhsffdeyj.supabase.co/functions/v1/reedle-mcp`

reedle