You can add OS-level automation to your AI IDEs with the Coasty MCP server. Drive real desktops, browsers, and terminals from Cursor or Claude without writing a single API call. This feature connects directly to the Coasty computer use API, letting you run task runs, workflows, and machine provisioning from your favorite AI assistant.
How it works
The MCP server exposes Coasty's core resources as tools your AI IDE can call. When you ask Cursor or Claude to automate a workflow, the assistant sends tool calls over the MCP transport. Coasty receives requests using the documented REST endpoints. For example, a task run is created with POST /v1/runs, which accepts machine_id, task, cua_version, and optional instructions, system_prompt, max_steps, deadline_seconds, and on_awaiting_human. The server drives an agent to completion and streams events via SSE until the run reaches a terminal state like succeeded or failed. You manage credits through a prepaid USD wallet where 1 credit equals $0.01. The MCP server authenticates using the COASTY_API_KEY environment variable and the X-API-Key header.
cat > ~/.config/claude/claude_desktop_config.json << 'EOF'
{
"mcpServers": {
"coasty": {
"command": "npx",
"args": ["-y", "@coasty/mcp-server"],
"env": {
"COASTY_API_KEY": "${COASTY_API_KEY}"
}
}
}
}
EOF
Key features exposed through MCP
- Create and manage task runs with POST /v1/runs, billed $0.05 per agent step
- Start, stop, and snapshot cloud machines via POST /v1/machines
- Define workflows with the versioned JSON DSL using POST /v1/workflows and POST /v1/workflows/{id}/runs
- Stream run events via GET /v1/runs/{id}/events with Last-Event-ID for reconnection
- Cancel or resume runs using POST /v1/runs/{id}/cancel and POST /v1/runs/{id}/resume
- Use cua_version 'v3' for guided control or 'v4' for autonomous runs with pass/fail verification
- Handle human approval states with on_awaiting_human ('pause', 'fail', or 'cancel')
Use the COASTY_API_KEY environment variable for authentication and the X-API-Key header in direct API calls.
Where this beats brittle automation
Traditional automation relies on brittle selectors that change layout or localization. A single UI shift breaks a script. The Coasty computer use API drives real desktops and browsers like a human. The MCP server lets your AI assistant reason about screen content and perform actions such as clicking, typing, and dragging based on visual context. This eliminates selector maintenance and lets you automate workflows that span multiple applications and windows.
Start automating desktop workflows from Cursor or Claude with the Coasty MCP server. Provision machines, run task workflows, and stream completion events directly from your AI IDE. Get your API key at https://coasty.ai/developers and build your first automated workflow today.
Want to see this in action?
View Case Studies