Pulumi MCP Server: Let AI Manage Infrastructure as Code

Why Give AI Access to Pulumi?

Pulumi lets you define infrastructure in real programming languages: TypeScript, Python, Go, C#. When you need to check stack state, preview changes, or deploy updates, you switch between your IDE and the Pulumi CLI or console. During incident response or rapid iteration, that context-switching adds up.

An MCP server for Pulumi lets your AI agent manage infrastructure directly. Preview deployments, check resource drift, query stack outputs, and trigger updates without leaving Claude, Cursor, or Windsurf.

What Becomes Possible

Once your AI agent has Pulumi access:

• "What resources are in the production stack?"
• "Preview the changes if I update the RDS instance class"
• "What's the current output of the networking stack?"
• "Are there any resources in a failed state?"
• "Deploy the staging stack with the latest config"
• "Show me the drift between desired and actual state"

How It Works

1. Go to DataFaucet and enter your Pulumi Cloud URL (e.g. https://app.pulumi.com)
2. Log into your Pulumi organization
3. Browse stacks, resources, and deployments as you normally would
4. Each action becomes an MCP tool your AI can call
5. Deploy and connect to Claude, Cursor, or Windsurf

Works with Pulumi Cloud and self-hosted backends. The AI agent gets whatever access your session provides.

Security Best Practices

• Read-only tokens for preview — create tokens scoped to preview and state reads only
• Environment isolation — separate tokens per stack environment (dev/staging/prod)
• No deploy tokens in dev — prevent accidental production deployments from AI agents
• Stack-level RBAC — use Pulumi Teams to restrict which stacks the AI can access
• Audit trail — Pulumi Cloud logs all operations with identity attribution

Use Cases by Role

Platform engineers: Stack management across environments. "What resources are in the prod-us-east stack?" without opening the console.

Backend developers: Infrastructure checks mid-coding. "What's the RDS endpoint output from the database stack?" without running pulumi commands.

SREs: Incident response. "Are there any resources in a pending or failed state?" for quick drift detection.

DevOps leads: Deployment oversight. "Show me all stacks that were updated in the last 24 hours" for change tracking.

Pulumi Console vs AI Agent

Getting Started

For read-only AI access, create a Pulumi access token with limited scope:

pulumi access-token create --scope stacks:read --description "AI agent read access"

Then point DataFaucet at your Pulumi console, authenticate, and your AI agent can query stack state and preview changes as MCP tools.