Pulumi MCP Server for Claude

Content

Pulumi MCP Server is Pulumi's Model Context Protocol support for connecting Claude and other MCP-capable assistants to Pulumi infrastructure workflows. The current Pulumi documentation focuses on the hosted remote MCP server at https://mcp.ai.pulumi.com/mcp, which uses OAuth and can connect an assistant to Pulumi Cloud resources, Registry context, policy findings, organization access, and Pulumi Neo tasks.

Use it when Claude needs current infrastructure context before writing or reviewing infrastructure as code: which stacks exist, what resources are deployed, which policies are failing, what Registry resources and properties are available, or whether a Pulumi Neo task should be launched. For first-time use, keep the assistant in inventory, documentation, and policy-review mode before delegating any infrastructure-changing work.

Pulumi also publishes the @pulumi/mcp-server npm package for local stdio workflows. That package documents Pulumi Registry tools and local Pulumi CLI tools, but the official docs now make the hosted OAuth endpoint the preferred setup path for most AI assistants.

Features

• Official Pulumi documentation page for MCP setup and supported assistants.
• Hosted remote MCP endpoint at https://mcp.ai.pulumi.com/mcp.
• OAuth authentication and organization selection through the MCP client.
• Claude Code, Cursor, Windsurf, Claude Desktop, and other MCP-compatible assistant setup paths.
• Pulumi Cloud stack and resource queries.
• Resource search across cloud infrastructure in a Pulumi organization.
• Pulumi Registry access for resource information, properties, examples, and documentation.
• Policy violation reporting for deployed infrastructure.
• Organization member and access management capabilities.
• Pulumi Neo task delegation for automated infrastructure planning and pull request workflows.
• Published @pulumi/mcp-server npm package for local stdio usage when a local package-based setup is required.

Tool Surface

Pulumi documents the hosted MCP server as a way for AI assistants to query Pulumi Cloud stacks and resources, search infrastructure across an organization, access Pulumi Registry information, get policy violation reports, manage organization members and access, delegate infrastructure tasks to Pulumi Neo, and generate infrastructure code using Registry tools and best practices.

The npm package README documents local tools such as Registry resource and function lookup, Pulumi CLI preview, update, refresh, stack output, resource search, and a Neo task launcher. Treat the local package as an alternative runtime with a broader local-machine blast radius because it can use local Pulumi CLI configuration and project files.

Installation

Claude Code Remote Server

Add the hosted Pulumi MCP server:

claude mcp add --transport http pulumi https://mcp.ai.pulumi.com/mcp

Start a Claude Code session, open /mcp, and complete the Pulumi OAuth flow when prompted. Choose the intended Pulumi organization and confirm that the assistant sees only the resources needed for the task.

Remote MCP Config

Use this remote HTTP configuration for clients that accept MCP JSON config:

{
  "mcpServers": {
    "pulumi": {
      "transport": "http",
      "url": "https://mcp.ai.pulumi.com/mcp"
    }
  }
}

For Claude Desktop clients that do not connect directly to remote HTTP MCP servers, use Pulumi's documented mcp-remote path and follow the OAuth prompt.

Local Package Alternative

If your workflow needs the local stdio package instead of the hosted endpoint, Pulumi publishes:

npx @pulumi/mcp-server@latest stdio

Use the local package only in a workspace where Pulumi CLI credentials, project files, cloud provider credentials, and stack configuration are approved for assistant access.

Use Cases

• Ask Claude to list stacks with production in the name before writing a change plan.
• Search deployed resources across a Pulumi organization during an inventory or migration review.
• Review policy violations before assigning remediation work.
• Look up Pulumi Registry resource properties and examples while drafting TypeScript, Python, Go, .NET, Java, or YAML Pulumi code.
• Inspect stack outputs or resource IDs needed for a support handoff.
• Ask whether a proposed task is safe to delegate to Pulumi Neo, then launch Neo only after the target repository and stack are confirmed.
• Use local package tools to preview a Pulumi project after confirming the workspace, stack, backend, and credentials.

Safety Checklist

• Connect the hosted server to the smallest Pulumi organization and role that satisfies the current task.
• Prefer inventory, Registry lookup, and policy review before any task that can create infrastructure changes.
• Confirm stack, project, repository, branch, and environment before launching a Pulumi Neo task.
• Keep MCP approval prompts enabled for member access changes, Neo delegation, local CLI preview/update/refresh tools, and any deployment-adjacent action.
• Use pulumi preview and policy findings before approving pulumi up or any generated pull request.
• Do not mount local project directories into a Docker or stdio MCP runtime unless the assistant should read those files.
• Rotate Pulumi access tokens created for temporary MCP work.
• Document organization-specific naming, policy, cost, and approval rules in project instructions before asking Claude to act on infrastructure.

Source Links

• Pulumi MCP server docs: https://www.pulumi.com/docs/ai/mcp-server/
• Pulumi remote MCP announcement: https://www.pulumi.com/blog/remote-mcp-server/
• npm package: https://www.npmjs.com/package/@pulumi/mcp-server
• Pulumi Neo docs: https://www.pulumi.com/docs/pulumi-cloud/neo/
• Pulumi Registry: https://www.pulumi.com/registry/
• MCP specification: https://modelcontextprotocol.io/specification/2025-06-18/