Connect Pushary to Cursor, Claude, or Lovable via MCP. Get a push notification the moment your AI task completes. Stop tab-switching. Free for a month.
I was spending 10+ hours a day in Cursor. I'd kick off a long agent task, check Twitter for 'one minute,' and come back 40 minutes later. The agent finished 35 minutes ago. This happened every single day.
I built Pushary for push notifications (it's a full platform for SaaS/e-commerce). Then I realized, why am I not using this for myself? I wired up an MCP server, and now my phone buzzes the second any AI agent completes.
4 lines of JSON in your MCP config. Works with Cursor, Claude Desktop, Windsurf, Lovable and anything that speaks MCP. Zero code. Your phone buzzes when the task is done.
Pushary is actually a full push notification platform (80% open rates, HubSpot sync, Shopify recovery, the works). But today I wanted to share the thing that changed my own daily workflow. Would love your feedback.
Try it and tell me what you think. What other AI workflows would you want notifications for?
Beyond “task finished,” which agent states ended up being the real time-sinks (e.g., waiting for input/approval, errors, tool timeouts), and how are you detecting those states reliably across different MCP clients?
@curiouskitty Great question. This is exactly the rabbit hole I went down after the initial "task done" notification.
The biggest time-sinks in order:
Waiting for human input. The agent needs a yes/no decision and just sits there. Worse than task completion because a blocked agent is burning context doing nothing.
Errors and blockers. Build failures, type errors, missing env vars. The agent hits a wall and you only find out 20 minutes later.
Long-running tasks completing. The original problem.
For detection, the MCP server exposes three tools:
send_notification sends a one-way push alert. The agent calls this when it finishes a task or hits an error. Your phone buzzes.
ask_user_yes_no sends a push notification with a question. You tap yes or no on your phone.
wait_for_answer blocks the agent until your answer comes back. Up to 55 seconds, retryable, your answer sticks around for 10 minutes.
The key insight: the agent decides when it's blocked, not us. We don't try to detect agent states from the outside. The MCP tools give the AI a way to reach you. When Cursor's agent needs approval to delete a migration file, it calls ask_user_yes_no("Delete the old migration?"). Your phone buzzes. You tap "yes" from the couch. The agent gets the answer and keeps going. No context switch back to the IDE.
This works across Cursor, Claude Desktop, Windsurf, Lovable, and anything that speaks MCP because MCP is the common protocol. Same 4 lines of JSON config everywhere.
The human-in-the-loop piece was the real unlock. "Task done" notifications save you 5-10 minutes per task. Decision notifications remove entire round-trips to the IDE.
What kind of agent workflows are you running? Happy to help you wire it up.
Replies
Pushary
Pushary
Happy to hear thoughts and comments from you guys how you're getting notified with pushary :)
Product Hunt
Pushary
@curiouskitty Great question. This is exactly the rabbit hole I went down after the initial "task done" notification.
The biggest time-sinks in order:
Waiting for human input. The agent needs a yes/no decision and just sits there. Worse than task completion because a blocked agent is burning context doing nothing.
Errors and blockers. Build failures, type errors, missing env vars. The agent hits a wall and you only find out 20 minutes later.
Long-running tasks completing. The original problem.
For detection, the MCP server exposes three tools:
send_notification sends a one-way push alert. The agent calls this when it finishes a task or hits an error. Your phone buzzes.
ask_user_yes_no sends a push notification with a question. You tap yes or no on your phone.
wait_for_answer blocks the agent until your answer comes back. Up to 55 seconds, retryable, your answer sticks around for 10 minutes.
The key insight: the agent decides when it's blocked, not us. We don't try to detect agent states from the outside. The MCP tools give the AI a way to reach you. When Cursor's agent needs approval to delete a migration file, it calls ask_user_yes_no("Delete the old migration?"). Your phone buzzes. You tap "yes" from the couch. The agent gets the answer and keeps going. No context switch back to the IDE.
This works across Cursor, Claude Desktop, Windsurf, Lovable, and anything that speaks MCP because MCP is the common protocol. Same 4 lines of JSON config everywhere.
The human-in-the-loop piece was the real unlock. "Task done" notifications save you 5-10 minutes per task. Decision notifications remove entire round-trips to the IDE.
What kind of agent workflows are you running? Happy to help you wire it up.
jared.so
When claude asks for permissions, does pushary also notify you? Congrats on the launch!
Pushary
@mcarmonas Of course, you just have to add this in a skill/rule file and it asks you a yes/no question like a G.
You can really touch grass now and steer AI from your lock screen ;)
Hey Aadil, tried it and this is exactly what I was looking for. I'm veryyyy interested to see where this is going. 🔥
Pushary
@faresgpt Legend!