A HelmA package manager for Kubernetes. A Helm chart bundles resource templates and default values so you can configure and install an application. values file can look perfectly reasonable and still configure nothing. The YAML is valid, the indentation is fine, but the chart expects a different key.
That’s an easy mistake for a coding assistant to make. Knowing how Helm works doesn’t mean it knows the configuration of the particular chart version you’re using.
I’ve built artifacthub-mcp, an MCP server that lets your assistant look up package information on Artifact Hub, including a chart’s actual values.yaml, rather than relying on what the model remembers.
Looking up the chart before writing the YAML
Artifact Hub is a catalog of cloud native packages. If you’ve searched for a Helm chart, you’ve probably ended up there. It has package metadata, versions, documentation, and chart information that an assistant can use while working on your configuration.
Model Context ProtocolA protocol that lets AI applications connect to servers exposing tools, resources, and prompts. Here, the server provides tools for looking up Artifact Hub package data. gives the assistant a way to call tools exposed by a server. In this case, those tools search Artifact Hub and retrieve information about its packages.
The part I’m most interested in is what happens after finding a chart. The server can fetch its default values, retrieve its values schema when available, and inspect individual chart templates. That gives the assistant something concrete to check before suggesting a configuration.
For example, you could ask:
Find the prometheus-community/prometheus chart on Artifact Hub. Check its default values and any available values schema, then help me write a values file with persistent storage for the Prometheus server. Tell me which chart version you’re using and explain each setting you change.
This is an example prompt, not a recorded demo. The useful test is whether the assistant retrieves the chart information and bases its answer on that version, rather than jumping straight to writing YAML.
Following a setting into the template
Sometimes the defaults don’t tell you enough. You want to know what a value actually changes.
The MCP server has tools to list a chart’s templates and retrieve their source. An assistant can use those to follow a setting into the template that consumes it.
A follow-up prompt might be:
Inspect the template that configures storage for the Prometheus server. Explain how the persistence settings affect the resources it creates, and point out any conditions in the template.
The tools make this a sequence of lookups. The assistant first gets the package metadata. It uses the package ID and version to list templates, then requests a template by its exact name. It doesn’t have to guess the filename.
There is an important limit here. Reading a Helm template isn’t the same as rendering the chart with your values. Conditions, dependencies, and cluster capabilities can affect the result. I’d still run helm template to inspect the rendered manifests, and validate the configuration before deploying it.
Versions and upgrades
The server also exposes package versions and changelogs. Those are useful when the question is about an existing deployment rather than a new one.
You can ask the assistant to list the available versions of a chart, or check the published changelog between your current version and a proposed upgrade. Keep the version explicit when asking it to inspect values or templates. Defaults for the latest release may not describe the release you’re running.
There’s also a tool for retrieving Artifact Hub’s package security reports. That’s a way to inspect the report published there, not a guarantee that a chart or your deployment is safe. The assistant should distinguish between a report with no findings and a report that isn’t available.
Trying it out
Install the server with Homebrew or Cargo:
brew tap luxass/homebrew-tap
brew install artifacthub-mcpcargo install --locked artifacthub-mcpThen register it with your MCP client:
claude mcp add artifacthub artifacthub-mcpcodex mcp add artifacthub artifacthub-mcpThe artifacthub-mcp docs have configuration examples for other clients and the full tool list.
You can also choose which tools the server exposes. For a session focused on finding packages and checking versions:
artifacthub-mcp --tools search_packages,get_package,get_package_versionsAlternatively, use --exclude-tools to leave out specific tools. The two flags are mutually exclusive.
A source to check against
To use artifacthub-mcp, install the binary, connect it to your assistant, and ask it to look up the chart you’re working with.
The assistant can still misread documentation or suggest the wrong configuration. What this adds is access to the package data it needs to check its answer, and a way for you to see which chart version and settings that answer is based on.
You can find the setup instructions and tool list in the artifacthub-mcp docs. If you try it, I’d be interested in examples where the assistant still guesses instead of using the available tools.