Pulumi MCP Server – AI Infrastructure Automation
Pulumi's hosted Model Context Protocol server lets AI-powered development tools query Pulumi Cloud stacks, resources, policies, users, and Registry documentation. It also delegates complex infrastructure workflows to Pulumi Neo, subject to the user's Pulumi organization access and authentication.
Overview
The Pulumi MCP Server is Pulumi's hosted Model Context Protocol integration for AI coding assistants. It connects assistants to Pulumi Cloud, the Pulumi Registry, and Pulumi Neo so infrastructure questions and actions can be handled from an MCP-capable editor or desktop client. Pulumi documents the hosted endpoint as https://mcp.ai.pulumi.com/mcp and describes OAuth-based connection flow where the user enters a Pulumi Access Token and selects an organization.
What the MCP server enables
The server gives an AI agent structured tools for Pulumi infrastructure workflows. Pulumi Cloud tools include get-stacks for listing stacks, resource-search for searching managed resources with Lucene queries, get-policy-violations for policy-as-code results, and get-users for organization member information. Registry tools include get-type, get-resource, get-function, list-resources, and list-functions, which help agents retrieve provider schemas, resource properties, functions, examples, and documentation before generating infrastructure code.
Pulumi also exposes Neo delegation tools such as neo-bridge, neo-get-tasks, neo-continue-task, and neo-reset-conversation. These can launch and manage Pulumi Neo tasks for multi-step infrastructure automation, such as analyzing policy violations, generating code changes, or creating pull-request workflows when the organization has the required Neo access.
When to use it
Use this MCP server when an AI assistant needs live Pulumi context rather than static IaC knowledge. Practical examples include asking which production stacks exist, searching for all S3 buckets across an organization, checking policy violations, generating code from Registry schemas, or delegating broad remediation tasks to Neo. It is especially useful inside AI-assisted development tools such as Cursor, Claude Code, Claude Desktop, Windsurf, and other clients that support MCP with OAuth or can bridge to a remote MCP endpoint.
Connection and authentication
Pulumi's primary documented connection is a hosted HTTP MCP endpoint at https://mcp.ai.pulumi.com/mcp. Authentication happens through the MCP client's browser-based OAuth flow: the user provides a Pulumi Access Token, selects the organization to use, and returns to the AI assistant. Clients that only support local stdio can use mcp-remote to connect to the hosted endpoint, but the actual Pulumi MCP service remains the remote HTTP endpoint.
Key considerations
Access is governed by the Pulumi token and the selected organization's permissions. Resource search results, stack data, policy violations, user listings, and Neo task actions should be treated as sensitive infrastructure information. Neo tasks may be subject to organization access and limits, so teams should use least-privilege Pulumi access tokens, validate generated code before deployment, and keep MCP client configuration under normal workstation security controls.
Supported Transports
streamable_http
URL: https://mcp.ai.pulumi.com/mcp
Frequently Asked Questions
- When should an AI agent use the Pulumi MCP Server?
- Use it when an AI assistant needs live Pulumi Cloud context or Pulumi Registry information, such as listing stacks, searching deployed resources, checking policy violations, or looking up resource schemas before generating infrastructure code.
- What does the Pulumi MCP Server add to an AI agent's capabilities?
- It adds structured access to Pulumi Cloud tools, Pulumi Registry lookups, and Pulumi Neo delegation workflows. This lets the agent answer questions about real infrastructure and launch supported infrastructure automation instead of relying only on model training data.
- What can the AI agent access or manage through this server?
- The documented tools include stack listing, organization-wide resource search, policy violation reports, organization users, Registry resource and function metadata, and Pulumi Neo task creation or continuation. Actual access depends on the Pulumi Access Token, selected organization, and available Neo permissions.
- How is authentication configured for this MCP server?
- Pulumi documents an OAuth-based connection flow for the hosted MCP server. On first connection, a browser page opens where the user enters a Pulumi Access Token and selects which Pulumi organization the assistant should use.
- Which transport should be used for the Pulumi MCP Server?
- Use the hosted HTTP MCP endpoint at `https://mcp.ai.pulumi.com/mcp` for clients that support remote MCP servers. For clients that only support stdio, Pulumi documents using `mcp-remote` as a bridge to that same hosted endpoint rather than a separate local Pulumi server.