entrydesk-cli

EntryDesk CLI

EntryDesk is a workspace for AI agents — run tasks against models, build and schedule agents, and reach connected SaaS tools. The entrydesk CLI exposes all of it from the terminal, so an agent can drive EntryDesk on the user's behalf. (A "task" is one AI conversation/run; the CLI verbs are task and tasks.)

When this skill applies

• "Ask EntryDesk …", "run this on on EntryDesk", "run my EntryDesk agent"
• The user pastes an EntryDesk share URL (https://<host>/share/<id>) and wants its contents
• "Search Slack / GitHub / Drive via EntryDesk", "what tools does my workspace have"
• "Schedule an agent to run daily"

Setup

npm install -g @entrydesk/cli
entrydesk login            # OAuth device flow: prints a URL + code to enter
entrydesk status           # confirm you're signed in
entrydesk workspaces       # list workspaces; switch with `workspaces switch <id>`

entrydesk login uses the OAuth device flow — it shows a verification URL and a user code; open the URL (the CLI tries to launch a browser, but you can also paste it anywhere) and enter the code. Most commands accept --json for scriptable output — prefer it when parsing.

Core actions

Run a task with a model or agent

entrydesk models --json                       # discover model ids
entrydesk task -m "Summarize this" --model <model-id>
entrydesk agents --json                        # discover agent ids
entrydesk task -m "Draft a reply" -a <agent-id>
entrydesk task -c <n> -m "follow up"           # continue task #n
entrydesk task -i                              # interactive mode
echo "explain this code" | entrydesk task --model <model-id>   # pipe stdin

Output: --plain for text only, --output stream-json for streaming structured events. List or view past tasks with entrydesk tasks / entrydesk tasks <n|uuid>.

Opening a shared task

When the user gives a URL like https://app.entrydesk.com/share/<id> (or any …/share/<id>, including custom domains):

# Extract the id after /share/ and pass it to --share (or -s):
entrydesk tasks --share <id>

Do NOT WebFetch / curl a share URL — those endpoints require auth and return 401. Always go through entrydesk tasks --share.

SaaS tools

EntryDesk fronts many SaaS services as callable tools. Discover, then call:

entrydesk tool list                       # all tools
entrydesk tool list --prefix slack        # filter by service
entrydesk tool get <name>                 # see a tool's input schema
entrydesk tool call <name> --input '{"key": "value"}'

Common service prefixes: slack, github, google_drive, google_gmail, google_calendar, ms365, grafana, redash. Run tool list --prefix <svc> to see the exact tool names — read each tool's schema via tool get before calling; don't assume parameter shapes.

Example — search Slack and read a message thread:

entrydesk tool call slack__search_all --input '{"query": "release notes"}'
entrydesk tool call slack__get_channel_info --input '{"channel": "C123"}'

Schedule an agent

entrydesk schedules create \
  --name "Daily Report" --agent-id <agent-id> \
  --prompt "Generate today's summary" \
  --starts-at "2026-06-01T09:00:00Z" \
  --repeat-type days --every 1 --time 09:00 --utc-offset "+08:00"
entrydesk schedules            # list
entrydesk schedules delete <id>

Pitfalls

• Login is device-flow only — entrydesk login prints a URL + code; there's no --password flag. It works headless (copy the URL/code to a browser elsewhere), but it does need a human to approve the code. Check entrydesk status before relying on auth.
• Wrong workspace — agents, models, and tools are workspace-scoped. If something "isn't there", confirm entrydesk workspaces shows the active one you expect; switch with workspaces switch <id>.
• Tool input schemas vary — always entrydesk tool get <name> first; the --input JSON must match that tool's schema or the call rejects.
• Discover ids, don't guess — model ids and agent ids come from entrydesk models --json / entrydesk agents --json. Hardcoding a stale id fails.
• Self-document via the CLI — entrydesk <command> --help is authoritative for flags; prefer it over assumptions when a command's options aren't obvious.