Drive Coasty from Cursor and Claude with the MCP Server
Most desktop automation lives in scripts that struggle with changing UIs and need brittle selectors. Coasty uses vision to see the screen and act like a human. The MCP server lets you call Coasty directly from Cursor, Claude Desktop, or any MCP client. This turns your IDE into a control room for real browsers and desktops.
How it works
The MCP server exposes Coasty as tools. Your IDE sends a tool call with a task description. The server calls POST /v1/predict with a base64 screenshot and instruction, then returns actions. It loops capture → predict → act until status is done. You can also use POST /v1/runs to have the server drive an agent to completion with a verifier. POST /v1/workflows lets you define a versioned JSON DSL of steps including tasks, asserts, loops, and human approvals.
curl https://coasty.ai/v1/predict \
-H "X-API-Key: $COASTY_API_KEY" \
-H "Content-Type: application/json" \
-d '{
"screenshot": "iVBORw0KGgoAAAANSUhEUgAAAAEAAAABCAYAAAAfFcSJAAAADUlEQVR42mNk+M9QDwADhgGAWjR9awAAAABJRU5ErkJggg==",
"instruction": "Click the Accept button.",
"cua_version": "v3"
}'
Configure the MCP server
- ●Install the Coasty MCP server from your package manager.
- ●Add its config to your IDE's MCP settings with the base URL https://coasty.ai/v1 and COASTY_API_KEY.
- ●Select the server in Cursor or Claude Desktop to enable the tools.
Use the MCP server to call Coasty directly from your IDE, not a separate script.
Where this beats brittle automation
Vision lets agents see the actual UI. They handle layout changes, dynamic elements, and brand variations without CSS selectors. They can click, type, scroll, and handle alerts just like a human. This is far more robust than XPath or ID-based automation that breaks when screens update. The MCP server keeps the agent in the same context as your code, so debugging flows, onboarding scripts, and interactive demos become natural.
Add computer use to Cursor or Claude Desktop with the MCP server. Build debugging flows that open browsers, run tasks, and report back. Get a key at https://coasty.ai/developers and start driving real desktops from your IDE.