mcp/pulumi repository overview

⁠Pulumi MCP Server

Pulumi MCP server enables AI-powered coding assistants to help you codify cloud architectures and get diffs for infrastructure changes right in your development environment.

What is an MCP Server?⁠

⁠Characteristics

⁠Available Tools (12)

⁠Tools Details

⁠Tool: deploy-to-aws

Deploy application code to AWS by generating Pulumi infrastructure. This tool automatically analyzes your application files and provisions the appropriate AWS resources (S3, Lambda, EC2, etc.) based on what it finds. No prior analysis needed - just invoke directly.

⁠Tool: neo-task-launcher

Launch a Neo task when user asks Neo to perform a task. Pulumi Neo is a purpose-built cloud infrastructure automation agent.

⁠Tool: pulumi-cli-preview

Run pulumi preview for a given project and stack

⁠Tool: pulumi-cli-refresh

Run pulumi refresh for a given project and stack

⁠Tool: pulumi-cli-stack-output

Get the output value(s) of a given stack

⁠Tool: pulumi-cli-up

Run pulumi up for a given project and stack

⁠Tool: pulumi-registry-get-function

Returns information about a Pulumi Registry function

⁠Tool: pulumi-registry-get-resource

Returns information about a Pulumi Registry resource

⁠Tool: pulumi-registry-get-type

Get the JSON schema for a specific JSON schema type reference

⁠Tool: pulumi-registry-list-functions

List all function types for a given provider and module

⁠Tool: pulumi-registry-list-resources

List all resource types for a given provider and module

⁠Tool: pulumi-resource-search

Search and analyze Pulumi-managed cloud resources using a strict subset of Lucene query syntax.

QUERY SYNTAX RULES:

• The search query syntax is a strict subset of Lucene query syntax
• The documents being searched are Pulumi resources
• The implicit operator is AND
• Parentheses and OR are supported between fields but not within fields
• All resources are returned by default (use empty query "" to get all)
• Wildcard queries are NOT supported (no * allowed)
• Fuzzy queries are NOT supported
• Boosting is NOT supported
• Field grouping is NOT supported
• Whitespace is NOT supported
• field:value produces a match_phrase query
• field:"value" produces a term query
• -field:value produces a bool must_not match_phrase query
• -field:"value" produces a bool must_not term query
• field: produces an existence query
• Resource properties can be queried with leading dot: .property.path:value or .property.path: (existence)
• You absolutely must not produce queries that use fields other than: type, name, id, stack, project, package, modified, provider, provider_urn, team and protected, unless the field is the name of a property.
• You absolutely must not produce queries that use wildcards (e.g., *).
• You absolutely must not produce queries that use field grouping (e.g., type:(a OR b))

AVAILABLE FIELDS:

• type: Pulumi types used for pulumi import operations (e.g., aws:s3/bucket:Bucket)
• name: logical Pulumi resource names
• id: physical Pulumi resource names
• stack: name of the stack the resource belongs to
• project: name of the project the resource belongs to
• created: when the resource was first created (absolute dates only)
• modified: when the resource was last modified (absolute dates only)
• package: package of the resource (e.g., aws, gcp)
• provider: alias for the "package" field
• provider_urn: full URN of the resource's provider
• protected: boolean representing whether a resource is protected
• team: name of a team with access to the resource

IMPORTANT QUERY PATTERNS: For AWS resources, do not use specific provider prefixes (aws: or aws-native:) in type filters. Instead: WRONG: type:aws:s3/bucket:Bucket WRONG: type:aws-native:s3:Bucket CORRECT: type:"Bucket" (searches across both aws and aws-native providers) For package filtering, use the generic package name: CORRECT: package:aws (matches both aws and aws-native packages) For finding resources by service, prefer the module field when possible: PREFERRED: module:s3 (finds all S3 resources regardless of provider) For property existence queries, always use the dot notation: CORRECT: .tags: (checks if tags property exists) For property negation queries (finding resources WITHOUT a property): CORRECT: -.tags: or NOT .tags: (finds resources without tags) COMMON TRANSLATIONS:

• "untagged resources" → -.tags: or NOT .tags:
• "resources without tags" → -.tags: or NOT .tags:

Supports field filters, boolean operators (AND, OR, NOT), exact matches with quotes, and property searches. The top parameter controls the maximum number of results to return (defaults to 20).

Resources may not have a repository url. This means that there is no available information about the repository that the resource is associated with.

⁠Use this MCP Server

{
 "mcpServers": {
 "pulumi": {
 "command": "docker",
 "args": [
 "run",
 "-i",
 "--rm",
 "-e",
 "PULUMI_ACCESS_TOKEN",
 "mcp/pulumi",
 "stdio"
 ],
 "env": {
 "PULUMI_ACCESS_TOKEN": "<PULUMI_ACCESS_TOKEN>"
 }
 }
 }
}

Why is it safer to run MCP Servers with Docker?⁠

Tag summary

This week's pulls