Model Context Protocol (MCP) is the open standard that lets AI agents connect to external systems securely. The Skyversal MCP server gives your agent the ability to list projects, read logs, roll back deployments and manage environment variables — all from the chat window.
🤖 What Is MCP and Why Use It?
Normally, checking a deployment means opening the console, finding the project and switching tabs. With MCP you simply ask your agent: "what's the status of aurora-launch's latest deploy?" — the agent pulls live data through the MCP server and summarises it. Zero context-switching.
📦 Setup (3 Steps)
1. Download the binary. Pick your platform from the table at skyversal.com/mcp. On macOS/Linux run chmod +x afterwards.
2. Log in via the CLI. The MCP server shares the Skyversal CLI credentials. Once, in your terminal:
skyversal login
3. Add it to your AI client. Point it at the downloaded file:
{
"mcpServers": {
"skyversal": {
"command": "/Users/YOU/skyversal-mcp-darwin-arm64"
}
}
}
Claude Desktop: claude_desktop_config.json · Cursor: Settings → MCP · Qoder: Settings → MCP Servers. The same block works everywhere.
🧰 Tool Catalog
Read-only: whoami, projects_list, project_get, deployments_list, deployment_logs, env_list.
State-changing (confirmation-gated): project_create, deploy_project, env_set, env_unset, rollback_deployment, cancel_deployment. The agent explains the effect and asks for your explicit confirmation before calling any of them (Human-in-the-Loop).
🛡️ Security & Limits
- ✓ Auth: uses the CLI Bearer token; nothing works without a session.
- ✓ Scope: only projects in your own workspace; foreign projects return 404.
- ✓ Local file upload is not possible over MCP; use
skyversal deployin your terminal for new code.
❓ Troubleshooting
- › "not logged in": run
skyversal loginin your terminal. - › Tools not showing: double-check the binary path and .exe extension in your config.
- › macOS "cannot be opened": allow it in System Settings → Privacy & Security.