use-latest-version-mcp¶
An MCP server that keeps an AI coding assistant from suggesting stale package versions out of its training data. It looks up the current version of any package — across 39 registries — and analyzes, plans, and (optionally) applies dependency upgrades for a whole project.
Why it exists¶
A model's training data has a cutoff. Left to itself it will confidently write
express@4.18.0 or requests==2.28.0 long after those are old. This server
gives the model live ground truth: it asks the registry, every time.
What it does¶
- Look up the latest version and metadata for a package on any supported registry, generate the exact install command, and compare against a version you already have.
- Scan a project for every dependency manifest and lock file, then report what is outdated with an upgrade type (major/minor/patch) and risk.
- Reason about versions: detect conflicting constraints, check whether a version satisfies a set of constraints, find the highest version that does, and build a step-by-step upgrade path through intermediate majors.
- Optimize -> validate -> apply: produce a whole-project upgrade plan, validate it for breaking changes and cycles, and write it to the manifests — dry-run by default, with automatic timestamped backups when it does write.
Highlights¶
- 39 registries, one consistent interface. See Supported Registries.
- 15 tools. See the Tool Reference.
- Local-first. No API key required. An optional
GITHUB_TOKENonly raises GitHub's rate limit for thegithub/ghcrregistries. - Resilient. Per-registry circuit breaker, request timeouts, response caching, and retries with backoff.
- Safe writes. The only tool that modifies files (
apply_upgrades) defaults to a dry run and backs up every file it touches.
Start with Getting Started.