Pulumi MCP Server

The Pulumi MCP Server connects AI assistants such as Claude Code, Codex, or Cursor to Pulumi Cloud and the Pulumi Registry. Anyone managing infrastructure with Pulumi can ask the assistant questions like "Which stacks exist in our organization?", "Where are unencrypted buckets running?", or "What does the resource for a managed Kubernetes cluster on AWS look like?" without switching between terminal, web console, and documentation. The server is operated by Pulumi itself. According to the documentation it is reachable as a hosted endpoint at mcp.ai.pulumi.com, so nothing needs to be installed locally.

What is the Pulumi MCP Server?

Pulumi describes cloud infrastructure in programming languages instead of plain configuration files. Pulumi Cloud manages the state of that infrastructure, the so-called stacks, along with the related policies and members. The MCP server makes this data queryable for language models and additionally bridges to the Pulumi Registry, where all available resources, functions, and types of the cloud providers are documented. The connection runs over Streamable HTTP with OAuth in the browser or, for automated environments, via a token in the header.

Which tools are available?

The documentation groups the tools into three areas. For Pulumi Cloud there is get-stacks to list the stacks of an organization, resource-search to search resources with a query syntax, get-policy-violations to show policy violations, and get-users for members and their roles. The Neo integration comprises neo-bridge to start tasks and approve follow-up requests, neo-get-tasks, neo-continue-task, and neo-reset-conversation. From the registry come list-resources, list-functions, get-resource, get-function, and get-type; they give the assistant the exact schemas so that it writes correct Pulumi code. In addition there is deploy-to-aws, which returns step-by-step instructions for deployments on AWS.

What is it particularly useful for?

The biggest benefit lies in two places. First, when writing infrastructure code: because the assistant queries the registry schemas directly, it no longer guesses properties and types from memory but looks them up. That noticeably reduces faulty programs. Second, in operations: platform and security teams can search inventories, list policy violations, and get an overview of stacks without building their own scripts. Anyone using Pulumi Neo, the vendor's AI agent, can start tasks through the server and approve follow-up questions in the chat.

Setup

Register https://mcp.ai.pulumi.com/mcp as a remote server. In Claude Code, claude mcp add --transport http pulumi https://mcp.ai.pulumi.com/mcp is enough. According to the documentation, clients without direct remote support can use npx -y mcp-remote as a bridge. On first connection you enter a Pulumi access token in the browser and choose the organization; the session lasts 30 days. For CI environments without a browser, the token can be passed as a bearer header together with the X-Pulumi-Org header. Use a dedicated token for this that is not identical to your CLI token.

Security and permissions

The assistant works with the identity of the token. According to the documentation, role-based access control, token expiry policies, and organization policies continue to apply. In practice that means: give the token only the rights the assistant really needs, and prefer a read-oriented token as long as you use the server only for lookups and analysis. Also keep in mind that responses containing stack and resource data enter the AI client context and therefore reach the model provider.

Costs

According to the documentation, the registry tools and deploy-to-aws work with any Pulumi account. Resource search and further features require one of the paid plans Essentials, Pro, or Enterprise, and Neo usage is billed separately. Pulumi lists a free plan without a credit card on its pricing page; which features are included in which plan is shown on the official pricing page.

Who is the Pulumi MCP Server for?

It is worthwhile for teams that already manage their infrastructure with Pulumi and want to use the AI assistant for writing, reviewing, and searching. Anyone using Terraform, OpenTofu, or plain cloud consoles gains nothing from it, because the server only knows Pulumi stacks and the Pulumi Registry. Beginners benefit from the registry tools with a free account, larger organizations mainly from resource search and policy checks.

A Pulumi Cloud account with an access token and an MCP client that supports remote servers (Streamable HTTP), such as Claude Code, Codex, Cursor, or VS Code. Resource search and Neo require a paid plan.

With the Pulumi MCP Server, an AI assistant queries stacks, resources, and policy violations of your Pulumi organization and looks up registry schemas. This guide leads from creating the token to the first query.

Step 1: Check the prerequisites

You need a Pulumi Cloud account, an access token, and an MCP client that supports remote servers via Streamable HTTP, such as Claude Code, Codex, or Cursor. According to the documentation, resource search and Neo require a paid plan.

Step 2: Create a token

Create a dedicated access token for the MCP server in Pulumi Cloud. According to the documentation it should not be identical to the token of your CLI. Choose an organization token or a personal token with the narrowest rights possible.

Step 3: Register the server

Register https://mcp.ai.pulumi.com/mcp as a remote server. In Claude Code, claude mcp add --transport http pulumi https://mcp.ai.pulumi.com/mcp is enough. Clients without remote support use npx -y mcp-remote https://mcp.ai.pulumi.com/mcp as a bridge.

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

Step 4: Sign in and choose an organization

On first connection a browser window opens. Enter the access token there and choose the organization. According to the documentation, the session stays valid for 30 days. In CI environments, pass the token as a bearer header together with the X-Pulumi-Org header.

Step 5: First queries

Start read-only: "List the stacks of my organization" or "Show the properties of the resource for an S3 bucket in the registry". Check that the answers belong to the right organization.

Use Neo with care

Through neo-bridge, Neo tasks can be started and follow-up questions approved. Neo can change infrastructure and is billed separately. Therefore always confirm approvals deliberately one by one.

OAuth flow in the browser with a Pulumi access token (organization or personal token, not the CLI token) and organization selection; alternatively bearer header plus X-Pulumi-Org for CI.

Access to stacks, resources, policy violations, and members of the selected organization as well as to the Pulumi Registry and Neo tasks, within the rights of the token used.

The server runs hosted by Pulumi. Stack, resource, and member data the assistant retrieves enter the AI client context and therefore reach the model provider.

Exposure of infrastructure data and member lists to the model provider, overly broad token rights, and unintentionally started or approved Neo tasks that can change infrastructure.