# SendCrew > LinkedIn actions for AI agents across many accounts, with a server-side safety layer the agent can't override. ## Connect - MCP (Streamable HTTP): https://sendcrew.io/api/mcp with header `Authorization: Bearer sc_live_...` - REST: https://sendcrew.io/api/v1 (same operations), OpenAPI at https://sendcrew.io/api/v1/openapi.json - Keys are per workspace (one client = one workspace). Create them in the dashboard under API keys. ## How writes work - Every write returns an `action_id` right away and runs later in that sender's queue, one action at a time. - Omit `sender_id` and SendCrew picks the healthiest warm sender with quota left; replies always go from the thread's owner. - `dry_run: true` checks a write against the safety layer without queueing it. - `idempotency_key` makes retries safe; `campaign` groups actions for stats and the campaign view. - Poll `get_action_status` or subscribe to webhooks. ## Safety envelope (server-side, not overridable) - Daily caps when warm: invites 25, messages 50, inmails 10, profileViews 80, engagements 30, withdrawals 20; invites 100/week. - Warmup starts at invites 5, messages 15, inmails 2, profileViews 20, engagements 8, withdrawals 10 and ramps over 14–21 days. - Working hours 8:00–19:00 in the sender's timezone; weekends at half caps; 45–180s between actions. - Never contacts the same person from two senders in a workspace. Checkpoints, captchas, restrictions and expired sessions pause the sender; nothing retries through them. - A blocked write returns HTTP 422 / MCP isError with `error: SAFETY_BLOCKED` and `details.reason`. Over-cap writes are not refused: they queue with `deferred_reason` and `run_after`. ## Tools - `get_profile(url, sender_id?, max_age_hours?)` [read] GET /api/v1/profile: Look up a LinkedIn profile by URL. Returns the stored profile if fresh; otherwise queues a profile view (counted against the sender's view cap) and returns an action_id to poll with get_action_status. - `send_connection(url, note?, sender_id?, dry_run?, idempotency_key?, campaign?)` [write] POST /api/v1/actions/send_connection: Send a connection request (optional note, max 300 chars). Auto-chains a profile view with a natural pause first. Writes are queued (never block on LinkedIn) and pass the server-side safety layer: daily/weekly caps, warmup, quiet hours in the sender's timezone, jitter and pool-wide dedupe. Omit sender_id to let SendCrew pick the healthiest warm sender with quota left. Set dry_run to check without queueing. - `send_message(url_or_thread, text, sender_id?, dry_run?, idempotency_key?, campaign?)` [write] POST /api/v1/actions/send_message: Message a 1st-degree connection by profile URL, or continue a thread by thread_id. Always sent from the sender that owns the contact or thread. Writes are queued (never block on LinkedIn) and pass the server-side safety layer: daily/weekly caps, warmup, quiet hours in the sender's timezone, jitter and pool-wide dedupe. Omit sender_id to let SendCrew pick the healthiest warm sender with quota left. Set dry_run to check without queueing. - `send_voice_note(url_or_thread, audio_url, sender_id?, dry_run?, idempotency_key?, campaign?)` [write] POST /api/v1/actions/send_voice_note: Send a voice note to a connection or thread. audio_url: a public https link to an .m4a or .mp3, at most 60 seconds and 10 MB. Writes are queued (never block on LinkedIn) and pass the server-side safety layer: daily/weekly caps, warmup, quiet hours in the sender's timezone, jitter and pool-wide dedupe. Omit sender_id to let SendCrew pick the healthiest warm sender with quota left. Set dry_run to check without queueing. - `send_inmail(url, subject, text, sender_id?, dry_run?, idempotency_key?, campaign?)` [write] POST /api/v1/actions/send_inmail: Send an InMail. Needs a Premium or Sales Navigator sender. Writes are queued (never block on LinkedIn) and pass the server-side safety layer: daily/weekly caps, warmup, quiet hours in the sender's timezone, jitter and pool-wide dedupe. Omit sender_id to let SendCrew pick the healthiest warm sender with quota left. Set dry_run to check without queueing. - `view_profile(url, sender_id?, dry_run?, idempotency_key?, campaign?)` [write] POST /api/v1/actions/view_profile: Visit a profile (a light warmup touch before connecting). Writes are queued (never block on LinkedIn) and pass the server-side safety layer: daily/weekly caps, warmup, quiet hours in the sender's timezone, jitter and pool-wide dedupe. Omit sender_id to let SendCrew pick the healthiest warm sender with quota left. Set dry_run to check without queueing. - `react_to_post(post_url, reaction?, sender_id?, dry_run?, idempotency_key?, campaign?)` [write] POST /api/v1/actions/react_to_post: React to a LinkedIn post (LIKE by default). Writes are queued (never block on LinkedIn) and pass the server-side safety layer: daily/weekly caps, warmup, quiet hours in the sender's timezone, jitter and pool-wide dedupe. Omit sender_id to let SendCrew pick the healthiest warm sender with quota left. Set dry_run to check without queueing. - `comment_on_post(post_url, text, sender_id?, dry_run?, idempotency_key?, campaign?)` [write] POST /api/v1/actions/comment_on_post: Comment on a LinkedIn post. Writes are queued (never block on LinkedIn) and pass the server-side safety layer: daily/weekly caps, warmup, quiet hours in the sender's timezone, jitter and pool-wide dedupe. Omit sender_id to let SendCrew pick the healthiest warm sender with quota left. Set dry_run to check without queueing. - `withdraw_invite(url, sender_id?, dry_run?, idempotency_key?, campaign?)` [write] POST /api/v1/actions/withdraw_invite: Withdraw a pending connection request (pool hygiene; stale invites are also withdrawn automatically after 21 days). Writes are queued (never block on LinkedIn) and pass the server-side safety layer: daily/weekly caps, warmup, quiet hours in the sender's timezone, jitter and pool-wide dedupe. Omit sender_id to let SendCrew pick the healthiest warm sender with quota left. Set dry_run to check without queueing. - `schedule_action(action, at_or_after, condition?)` [write] POST /api/v1/actions/schedule: Queue any write to run at or after a time, optionally only if a condition still holds then: if_connected (they accepted), if_no_reply (no reply since scheduling), if_not_replied_by (no reply by a date). Follow-ups without a sequence builder. Writes are queued (never block on LinkedIn) and pass the server-side safety layer: daily/weekly caps, warmup, quiet hours in the sender's timezone, jitter and pool-wide dedupe. Omit sender_id to let SendCrew pick the healthiest warm sender with quota left. Set dry_run to check without queueing. - `check_inbox(since?, sender_id?, unread_only?, tag?, cursor?, limit?, include_messages?)` [read] GET /api/v1/inbox: Recent conversations across every sender in the workspace (or one sender), newest first, with the last messages of each. - `get_thread(thread_id)` [read] GET /api/v1/threads/{thread_id}: Full message history of one thread. - `reply_to_thread(thread_id, text, dry_run?, idempotency_key?)` [write] POST /api/v1/threads/{thread_id}/reply: Reply in a thread. Always sent from the sender that owns the thread. Writes are queued (never block on LinkedIn) and pass the server-side safety layer: daily/weekly caps, warmup, quiet hours in the sender's timezone, jitter and pool-wide dedupe. Omit sender_id to let SendCrew pick the healthiest warm sender with quota left. Set dry_run to check without queueing. - `tag_thread(thread_id, tags, mode?)` [write] POST /api/v1/threads/{thread_id}/tags: Label a thread (e.g. interested, not-now, booked). mode: set (default), add or remove. - `list_senders()` [read] GET /api/v1/senders: Senders in this workspace with status, health score, warmup stage and quota left today per action type. - `get_sender_health(sender_id)` [read] GET /api/v1/senders/{sender_id}: One sender's health score, warmup stage, quotas, proxy status and recent events (checkpoints, failures, accepts). - `get_action_status(action_id)` [read] GET /api/v1/actions/{action_id}: Status and result of a queued action (scheduled, queued, running, succeeded, failed, cancelled, skipped). - `list_actions(status?, sender_id?, campaign?, cursor?, limit?)` [read] GET /api/v1/actions: Recent actions, newest first, filterable by status, sender or campaign. Cursor-paginated. - `cancel_action(action_id)` [write] POST /api/v1/actions/{action_id}/cancel: Cancel an action that hasn't started yet. - `get_stats(range?, sender_id?, campaign?)` [read] GET /api/v1/stats: Acceptance and reply rates per sender and in total for a range (24h, 7d, 30d, 90d), optionally for one campaign. - `find_email(url)` [read] POST /api/v1/find_email: Find a work email for a profile. Costs 1 email credit only when an email is found; the result is kept only in this workspace. ## Webhook events - `message.received`: A contact replied in a thread. - `connection.accepted`: A contact accepted a connection request. - `action.succeeded`: A queued action ran successfully. - `action.failed`: A queued action failed. - `sender.health_changed`: A sender's health bucket changed (healthy, throttled, critical, stopped). - `sender.paused`: A sender was paused by a checkpoint, restriction or expired session. - `sender.resumed`: A paused sender is active again. Deliveries are signed: `SendCrew-Signature: t=,v1=." with your signing secret>`. Retries back off over ~15 hours.