Pulsar Deploy MCP Server

An MCP (Model Context Protocol) server that lets AI agents — Claude Desktop, Cline, Continue, Cursor — drive a Pulsar Deploy account directly. Ask Claude "deploy my-app" or "spin up a Postgres for the api project" and it just works.

Setup

1. Get a token. Sign in to https://pulsardeploy.com → Settings → Tokens → Create. Copy the pulsar_xxxxxxxx value.

2. Add to your MCP client config. Example for Claude Desktop (~/Library/Application Support/Claude/claude_desktop_config.json on macOS or %APPDATA%\Claude\claude_desktop_config.json on Windows):

   {
     "mcpServers": {
       "pulsar-deploy": {
         "command": "npx",
         "args": ["-y", "pulsar-deploy-mcp"],
         "env": {
           "PULSAR_API_TOKEN": "pulsar_xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx",
           "PULSAR_API_URL":   "https://api.pulsardeploy.com"
         }
       }
     }
   }

3. Restart your MCP client. Open a new chat and try:

   • "what projects do I have on pulsar?"
   • "deploy my-app"
   • "tail the logs for that deployment"
   • "create a postgres database called api-db and attach it to my-app"

PULSAR_API_URL is optional — defaults to the hosted Pulsar Deploy. Self-hosted users set it to their own API origin.

What's exposed

The server intentionally surfaces high-level intent rather than every REST route. 22 tools cover the day-to-day:

| Tool | What it does | | ---------------------------- | ------------------------------------------------------------------- | | list_projects | All your projects with live URL + last deploy status | | get_project | One project, full config + last 10 deploys | | create_project_from_github | Paste a GitHub URL — framework auto-detected, build config inferred | | delete_project | Destructive; confirm with user | | deploy | Trigger production deploy at HEAD | | get_deployment | Status, URL, duration, error | | tail_logs | Last N build log lines for a deployment | | search_logs | Substring-match across all stored build/deploy logs | | rollback | Roll back to previous live deploy | | list_env_vars | Masked — safe to show user | | set_env_var / delete_env_var | Encrypted server-side; effect on next deploy | | list_databases | Managed Postgres / MySQL / Redis / Mongo / PocketBase | | create_database | Auto-links DATABASE_URL (or equivalent) to a named project | | get_connection_string | Owner-only | | run_sql | Against a managed Postgres/MySQL — DROP DATABASE/ROLE blocked | | restart_database | Stop + start a managed instance | | list_domains / add_domain | Custom domains + DNS instructions | | wake_project | Wake the most recent sleeping deployment | | get_resource_stats | CPU / memory / network for a project's live container | | health_summary | One-shot status across every project |

The handler accepts project / database arguments as either UUIDs or names, so the AI doesn't need to round-trip via list_projects every time.

Self-hosting / development

cd packages/mcp-server
npm install
npm run dev   # tsx watch
# or
npm run build && node dist/index.js

Security

• Auth is your Personal Access Token. Treat it like a password — anyone with it can deploy / delete / read env vars in your account.
• The server runs locally in your MCP client's process. The token never leaves your machine except in the Authorization: Bearer ... header to your configured PULSAR_API_URL.
• delete_project and delete_env_var are flagged "destructive" in the tool descriptions — well-behaved AI clients will confirm with you first.