LaunchDeploy

Account

MCP for AI agents

One URL, every agent

LaunchDeploy is a remote MCP server at /mcp (Streamable HTTP). Add it to Cursor, Claude Code, VS Code, Windsurf, Codex or any MCP client and the agent can do everything the web app does: connect providers, create a project, review and start a launch, follow each step, fix DNS and email, manage environment variable keys, read the Launch Score and monitoring. Account → Security shows ready-to-paste configuration for each client, and an Add to Cursor button.

Authentication

The server uses the same personal access tokens as the API. Create one under Account → Security → API tokens and send it as Authorization: Bearer ld_… — in Cursor and VS Code that is the headers object of the server entry, in Claude Code the --header flag of claude mcp add, in Codex the bearer_token_env_var setting. The agent acts as you, inside the workspace the token was created in, and every change it makes appears in the activity log under your name.

Tools

get_workspace, list_providers, connect_provider, validate_provider, disconnect_provider, browse_provider · list_stack_presets, list_projects, get_project, create_project, update_project, archive_project · plan_launch, get_launch, start_launch, cancel_launch, rollback_launch, manage_launch_step, create_supabase_organization · get_domain, update_nameservers, get_dns, manage_dns_snapshots · get_email, setup_email, verify_email · list_deployments, redeploy · list_environment_variables, set_environment_variable, remove_environment_variable · get_health, fix_everything, propose_fix, apply_fix, ignore_issue · get_monitoring, update_alert, get_activity. The launch-project prompt walks an agent through the whole flow.

Guardrails

Tools carry MCP annotations (read-only, idempotent, destructive) so clients can ask before running the risky ones. Irreversible actions — rolling back a launch or step, restoring a DNS snapshot, moving nameservers, high-impact fixes — additionally require confirmed=true, which the server instructs agents to set only after you explicitly agreed. Environment variable values are never exposed to agents: the tools return key names only, and fetching values stays in the web app behind password confirmation. Provider credentials are never returned. The endpoint is rate limited like the API.

Try it

In your agent, ask: "Use LaunchDeploy to launch my repo acme/site on acme.com with the simple_website preset." The agent will list what to connect, ask you for provider tokens, create the project, show you the launch plan, start it, poll until it finishes and run the Production Doctor. Verify the connection from a shell with curl -X POST https://your-domain/mcp -H "Authorization: Bearer ld_…" -H "Accept: application/json, text/event-stream" -d '{"jsonrpc":"2.0","id":1,"method":"tools/list"}'.