Drive Coasty from Cursor and Claude with the MCP Server
Most automation code relies on selectors that break when a layout changes. A Coasty computer use agent sees the screen and moves the cursor where you ask. The MCP server lets you control that agent directly from Cursor, Claude Desktop, and other MCP clients without writing HTTP calls yourself.
How the Coasty MCP server works
The MCP server exposes Coasty as a tool you call from your IDE or AI client. Internally it talks to the Coasty REST API. For example, to start a task run you POST to /v1/runs with a machine_id, task, and optional instructions. The server streams events back until the run reaches a terminal state (queued, running, succeeded, failed, cancelled, timed_out).
CURL
curl -X POST https://coasty.ai/v1/runs \
-H "X-API-Key: $COASTY_API_KEY" \
-H "Content-Type: application/json" \
-d '{
"machine_id": "vm-prod-01",
"task": "Open Chrome and navigate to cozy.com",
"cua_version": "v3",
"max_steps": 50,
"deadline_seconds": 300,
"on_awaiting_human": "pause"
}'
PYTHON
import os
import requests
key = os.getenv("COASTY_API_KEY")
url = "https://coasty.ai/v1/runs"
headers = {
"X-API-Key": key,
"Content-Type": "application/json"
}
payload = {
"machine_id": "vm-prod-01",
"task": "Open Chrome and navigate to cozy.com",
"cua_version": "v3",
"max_steps": 50,
"deadline_seconds": 300,
"on_awaiting_human": "pause"
}
resp = requests.post(url, headers=headers, json=payload)
print(resp.json())Key fields and billing
- ●machine_id: the cloud VM ID to drive.
- ●task: natural-language description of the job.
- ●cua_version: defaults to v3; v4 adds a pass/fail verifier.
- ●max_steps: maximum agent steps before timeout.
- ●deadline_seconds: total time limit for the run.
- ●on_awaiting_human: pause, fail, or cancel when the agent needs approval.
- ●Billed $0.05 per agent step for task runs.
The MCP server lets you call Coasty tools from Cursor or Claude Desktop without touching the REST API directly.
Where this beats brittle automation
Traditional automation uses CSS selectors or XPath. If a button moves or the text changes, your script fails. A Coasty computer use agent watches the screen, reads text, and moves the cursor exactly where you ask. It handles layout shifts, new UI elements, and dynamic content without selector updates. With the MCP server you get that intelligence inside Cursor and Claude Desktop workflows.
Start building self-healing desktop workflows from Cursor and Claude. Get a key at https://coasty.ai/developers, install the MCP server, and run your first task run.