Use the local Notion CLI (notion-cli-agent) to query, create, update, and manage Notion pages and databases via shell. Use when interacting with Notion workspaces, querying databases, creating or updating pages, managing tasks, reading content blocks, or running bulk/batch operations on Notion data. Prefer over Notion MCP or API calls.
Local CLI for full Notion access.
notion <args> # globally installed via npm
Token resolution order inside the CLI (getTokenSync() in client.ts):
NOTION_TOKEN env var (highest priority)NOTION_API_KEY env var~/.config/notion/api_key~/.notion/tokenGotcha: Claude Code sessions often inherit a stale NOTION_TOKEN env var from a parent process (older sessions, MCP config, etc.). When that happens, notion <cmd> returns 401 API token is invalid even though ~/.config/notion/api_key contains a perfectly valid token — the CLI never reads the file because the env var wins.
Diagnostic: notion doctor will show ✅ Token: Token found but ❌ API Connection: Token is invalid or expired. That pattern = stale env var, NOT a bad file.
Reliable workaround (always works from Claude Code sessions):
NOTION_TOKEN=$(cat ~/.config/notion/api_key) notion <cmd>
This forces the CLI to use the file-backed token directly, bypassing the stale env var. Use this as the default invocation inside Claude sessions unless you've already verified notion doctor is fully green.
If you want to fix it permanently, the correct, source-of-truth token lives in ~/.config/notion/api_key. Update .zshrc, ~/.mcp/global-config.json, and any other places where NOTION_TOKEN is exported to match the file, then restart the session.
notion sync # cache all databases locally
notion list # show cached databases
notion list --json # for parsing
After sync, use database names instead of UUIDs in ALL commands:
notion db query "Tasks" --limit 5 --llm
notion find "Tasks" "overdue" --llm
notion stats overview "Projects"
If ~/.config/notion/workspace.json exists (from sync or the notion-onboarding skill), names resolve automatically. Falls back to UUIDs if no cache.
notion sync (once per session, or notion list if already synced)notion inspect context "Tasks" or notion inspect schema "Tasks" --llmdb query --title, search --exact --db --first, or --llm over fuzzy workspace-wide search when you know the target DB--dry-run first on bulk/batch ops, then confirm with usernotion sync # cache databases for name lookup
notion list # show cached databases
notion inspect ws --compact # all databases, names + ids
notion inspect schema "Tasks" --llm # property types + valid values
notion inspect context "Tasks" # workflow context + examples
notion ai prompt "Tasks" # DB-specific agent instructions
# Exact lookup in a known DB (deterministic — uses database query API)
notion db query "Tasks" --title "Known Page" --json
notion db query "Tasks" --limit 20 --llm # compact output
# Fuzzy search (workspace-wide, best-effort — Notion may miss long titles)
notion search "keyword" --limit 10
notion search "keyword" --db <db_id> --llm # filter by parent DB
notion search "short title" --exact --first --json # best-effort exact match
# Natural language
notion find "overdue tasks unassigned" -d <db_id> --llm
notion find "high priority" -d <db_id> --explain # preview filter, don't run
For exact lookup by title in a known DB, always use db query --title — not search --exact. Notion's search API is fuzzy and may miss pages with long or common-word titles.
notion page get <page_id> # properties
notion page get <page_id> --content # + content blocks
notion page get <page_id> --json # raw JSON
notion page read <page_id> # content as Markdown (native API, 1 call)
notion page read <page_id> --blocks # legacy block-by-block conversion
notion page read <page_id> -o page.md # save to file
notion ai summarize <page_id> # concise summary
notion ai extract <page_id> --schema "email,phone,date"
notion page create --parent <db_id> --title "Task Name"
notion page create --parent <db_id> --title "Task" --prop "Status:status=Todo" --prop "Priority:select=High"
notion page update <page_id> --prop "Status:status=Done"
notion page update <page_id> --clear-prop "Assignee" # type-aware clear
notion page update <page_id> --clear-prop "Tags" --clear-prop "Deadline"
notion page write <page_id> -f content.md # write Markdown to page
notion page write <page_id> -f doc.md --replace # replace all content
notion page edit <page_id> --at 3 --delete 2 # surgical block editing
notion page edit <page_id> --at 5 --markdown "New text" # insert at position
notion block append <page_id> --text "Paragraph"
notion block append <page_id> --heading2 "Section" --bullet "Item 1" --bullet "Item 2"
notion block append <page_id> --todo "Action item"
notion batch --dry-run --data '[
{"op":"get","type":"page","id":"<page_id>"},
{"op":"create","type":"page","parent":"<db_id>","data":{"title":"New"}},
{"op":"update","type":"page","id":"<page_id2>","data":{"Status":"Done"}}
]'
notion batch --llm --data '[...]' # execute
notion bulk update <db_id> --where "Status=Todo" --set "Status=In Progress" --dry-run
notion stats overview <db_id>
notion validate check <db_id> --check-dates --check-stale 30
notion dedup <db_id> # find duplicate pages
notion dedup <db_id> --fuzzy # include near-duplicates
notion dedup <db_id> --fix --strategy keep-largest --yes # archive duplicates
| Flag | Use for |
|---|---|
--llm | Compact, structured output for agents (search, db query, find, batch, inspect schema/context, stats overview, relations backlinks) |
--json / -j | Raw JSON for parsing |
--csv | CSV with headers (db query, find) |
--tsv | Tab-separated (db query, find) |
--ids-only | One ID per line for piping (db query, search, find) |
| (default) | Human-readable |
--filter-prop-type is required for non-text properties:
notion db query <db_id> \
--filter-prop "Status" --filter-type equals \
--filter-value "Done" --filter-prop-type status
Types: status · select · multi_select · number · date · checkbox · people · relation
See references/filters.md for full operator reference.
Auto-detection treats plain strings as select. Use Key:type=Value to force a type:
notion page update <id> --prop "Status:status=Done" # status, not select
notion page update <id> --prop "Notes:rich_text=Text" # rich_text, not select
notion page update <id> --prop "Owner:people=<user_id>" # people
inspect context0.10.0 (resolvePropertyName() is case-insensitive and whitespace-tolerant), but still prefer the real schema labels for reliability"Name", "Título", "Task" — check state or schema)db query --title "..." or search --db <id> --exact --first when you know the DB; avoid fuzzy search for operational updates--clear-prop instead of fake empty values like Owner:people= or Tags=--dry-run before any bulk/batch writereferences/filters.md — all property types × filter operators with examplesreferences/batch-patterns.md — batch workflows (multi-update, bulk status sweep, multi-get)references/workflows.md — agent workflow recipes (task triage, weekly review, project sync)notion quickstart # full quick reference
notion <command> --help # per-command help
notion ai suggest <db_id> "what I want to do"