Pulumi基础设施自动化
STDIOOfficial用于与Pulumi CLI交互的MCP服务器
用于与Pulumi CLI交互的MCP服务器
Note: This MCP server is currently under active development. Its API (including available commands and their arguments) is experimental and may introduce breaking changes without notice. Please file an issue on GitHub if you encounter bugs or need support for additional Pulumi commands.
A server implementing the Model Context Protocol (MCP) for interacting with Pulumi CLI using the Pulumi Automation API and Pulumi Cloud API.
This package allows MCP clients to perform Pulumi operations like retrieving package information, previewing changes, deploying updates, and retrieving stack outputs programmatically without needing the Pulumi CLI installed directly in the client environment.
The Pulumi CLI has to be installed on you machine.
This package is primarily intended to be integrated into applications that can use MCP servers as AI tools.
For Claude Code, you can install this MCP server using:
claude mcp add -s user pulumi -- npx @pulumi/mcp-server@latest stdio
For Claude Desktop, you can include Pulumi MCP Server in the MCP configuration file:
{ "mcpServers": { "pulumi": { "command": "npx", "args": ["@pulumi/mcp-server@latest","stdio"] } } }
Or if you prefer HTTP with Server-Sent Events (SSE) instead of stdio
:
{ "mcpServers": { "pulumi": { "command": "npx", "args": ["@pulumi/mcp-server@latest","sse"] } } }
You can also run the Pulumi MCP Server as a Docker container. This approach eliminates the need to install Node.js and the package dependencies directly on your host machine.
To build the container:
docker build -t pulumi/mcp-server:latest .
To use the containerized server with MCP clients, you'll need to configure the client to use the Docker container. For example, in Claude desktop's MCP configuration:
{ "mcpServers": { "pulumi": { "command": "docker", "args": ["run", "-i", "--rm", "pulumi/mcp-server:latest", "stdio"] } } }
To use the containerized server with MCP clients over HTTP (SSE), you can run the container with the following command:
{ "mcpServers": { "pulumi": { "command": "docker", "args": ["run", "-i", "--rm", "-p", "3000:3000", "pulumi/mcp-server:latest", "sse"] } } }
For Pulumi operations that require access to local Pulumi projects, you'll need to mount the appropriate directories. For example, if your Pulumi project is in ~/projects/my-pulumi-app
:
{ "mcpServers": { "pulumi": { "command": "docker", "args": ["run", "-i", "--rm", "-v", "~/projects/my-pulumi-app:/app/project", "pulumi/mcp-server:latest"] } } }
Then when using the MCP tools, you would reference the project directory as /app/project
in your requests.
The server exposes handlers for the following Pulumi operations, callable via MCP requests:
preview
: Runs pulumi preview
on a specified stack.
workDir
(string, required): The working directory containing the Pulumi.yaml
project file.stackName
(string, optional): The stack name to operate on (defaults to 'dev').up
: Runs pulumi up
to deploy changes for a specified stack.
workDir
(string, required): The working directory containing the Pulumi.yaml
project file.stackName
(string, optional): The stack name to operate on (defaults to 'dev').stack-output
: Retrieves outputs from a specified stack after a successful deployment.
workDir
(string, required): The working directory containing the Pulumi.yaml
project file.stackName
(string, optional): The stack name to retrieve outputs from (defaults to 'dev').outputName
(string, optional): The specific stack output name to retrieve. If omitted, all outputs for the stack are returned.get-resource
: Returns information about a specific Pulumi Registry resource, including its inputs and outputs.
provider
(string, required): The cloud provider (e.g., 'aws', 'azure', 'gcp', 'random') or github.com/org/repo
for Git-hosted components.module
(string, optional): The module to query (e.g., 's3', 'ec2', 'lambda').resource
(string, required): The resource type name (e.g., 'Bucket', 'Function', 'Instance').list-resources
: Lists available resources within a Pulumi provider package, optionally filtered by module.
provider
(string, required): The cloud provider (e.g., 'aws', 'azure', 'gcp', 'random') or github.com/org/repo
for Git-hosted components.module
(string, optional): The module to filter by (e.g., 's3', 'ec2', 'lambda').deploy-to-aws
: Deploy application code to AWS by generating Pulumi infrastructure. Automatically provisions AWS resources (S3, Lambda, EC2, etc.) based on application type.
The server also provides prompts that can be used by MCP clients:
deploy-to-aws
: AWS deployment guidance prompt that provides comprehensive instructions for analyzing application code and generating Pulumi infrastructure-as-code for AWS deployments. Includes best practices for security, cost optimization, and multi-environment setups.make ensure
make build
make test
For local testing using Claude Code, you can use the provided scripts:
# Build the project npm run build # Install the MCP server locally for testing ./scripts/install-mcp.sh # Or install from a packed .tgz file (created with npm pack) npm pack ./scripts/install-mcp.sh pulumi-mcp-server-0.1.2.tgz # Remove the local installation ./scripts/uninstall-mcp.sh
The install script will remove any existing pulumi-mcp-local
installation and install the current version from your local directory. Optionally, you can provide a .tgz
file created with npm pack
to test the packaged version. Run Claude Code and type the /mcp
command to see it.
Use MCP Inspector to list the existing tools and their metadata:
npx @modelcontextprotocol/inspector --cli node dist/index.js stdio --method tools/list
Use MCP Inspector to list the existing prompts and their metadata:
npx @modelcontextprotocol/inspector --cli node dist/index.js stdio --method prompts/list
This project is licensed under the Apache-2.0 License. See the LICENSE file for details.