package-version-check-mcp

A MCP server that returns the latest stable versions of packages you use as dependencies in a variety of ecosystems, such as Python, NPM, Go, or GitHub Actions.

It also supports looking up the latest versions of almost 1000 tools, such as development runtimes like python, node, dotnet, development tools like gradle, and various DevOps tools like kubectl or terraform, via the mise-en-place tool.

Why do I need this?

Whenever an AI coding agents generates files that pin dependency versions, they insert outdated versions because their training happened weeks or months ago, and new dependency versions have been released since then. As a developer, it is annoying having to manually fix these outdated versions.

This MCP fixes this problem. Use it together with an MCP such as Context7 to avoid that your AI agent produces outdated code.

Features

Supported ecosystems / tools:

• Developer ecosystems:
  • NPM - Node.js packages from registry.npmjs.org
  • PyPI - Python packages from PyPI
  • NuGet - .NET packages from NuGet
  • Maven / Gradle - Java/Kotlin/Scala packages from Maven repositories (Maven Central, Google Maven, etc.)
  • Go - Go modules from proxy.golang.org
  • PHP - PHP packages from Packagist (used by Composer)
  • Ruby - Ruby gems from rubygems.org
  • Rust - Rust crates from crates.io
  • Swift - Swift packages from GitHub repositories
  • Dart - Dart packages from pub.dev
• DevOps ecosystems:
  • Docker - Docker container images from Docker registries
  • Helm - Helm charts from ChartMuseum repositories and OCI registries
  • GitHub Actions - Actions hosted on GitHub.com, returning their current version, their inputs and outputs, and (optionally) their entire README with usage examples
  • Terraform Providers and Modules - Providers & Modules from Terraform Registry, OpenTofu Registry, or custom registries
  • Various tools such as kubectl, terraform, gradle, maven, etc. supported by mise-en-place

Usage

Adding the MCP to Your Agent

There are three ways to make this MCP available to your AI coding agent:

Option 1: Use the Hosted Service (Easiest)

Point your agent to the free hosted service:

https://package-version-check-mcp.onrender.com/mcp

in (streamable) HTTP mode.

This is the quickest way to get started. Note that the hosted service may have rate limits from the underlying package registries.

Option 2: Run with uvx (for local use)

Use uvx to run the MCP server locally:

uvx package-version-check-mcp --mode=stdio

This automatically installs and runs the latest version from PyPI.

Requirements:

• You need the mise binary on PATH if you want to call the tools get_supported_tools or get_latest_tool_versions

Optional but recommended: Set the GITHUB_PAT environment variable to a GitHub Personal Access Token (no scopes required) to avoid GitHub API rate limits.

Option 3: Run with Docker (for local use)

Use the pre-built Docker image:

docker run --pull=always --rm -i ghcr.io/mshekow/package-version-check-mcp:latest --mode=stdio

Optional but recommended: Pass the GITHUB_PAT environment variable using -e GITHUB_PAT=your_token_here to avoid GitHub API rate limits.

Caching Configuration

To improve performance and reduce API calls to package registries, you can enable caching:

• PACKAGE_VERSION_CACHE_ENABLED: Set to true to enable caching (disabled by default)
• PACKAGE_VERSION_CACHE_TTL_SECONDS: Cache duration in seconds (default: 3600 / 1 hour)
• PACKAGE_VERSION_CACHE_MAX_SIZE_MB: Maximum cache size in MB (default: 64)

The cache is an in-memory TTL (Time-To-Live) cache. It resets when the MCP server restarts.

Configuring Your Agent

Once you've added the MCP server, you need to:

1. Enable the MCP tools in your agent's configura

... View full README on GitHub