I created a tool but it's not exposed when I run `kubectl-ai` as an MCP server

[Mostafa-wael]

I created a Helm tool in the ~/.config/kubectl-ai/tools.yaml as stated in the documentation, I can see it when I run kubectl-ai tools like this

However, I can't find the' Helm' tool when I expose it as an MCP server using VScode.

Of course, I tried refreshing it multiple times and restarting the server.

Any help?

[mikebz]

I don't believe that adding a tool to kubectl-ai as a client is necessarily exposes it as an MCP server.

[Mostafa-wael]

Mmmm, I thought that exposing it as an MCP server will expose all its tools. Like, I use kubectl-ai heavily with Helm, so why can't I utilise it too through VS code.

Would something like this be considered in the future?

[droot]

Thanks for reporting it @Mostafa-wael . I remember seeing another issue reporting the same problem.

Like, I use kubectl-ai heavily with Helm, so why can't I utilise it too through VS code.

Wow, can you share more about your usage. We are collecting some data points, and it will help us to understand how end-users are using the tool.

Confirming, as of today, we don't expose the custom tools (configured through mcp-client) through the mcp-server. I think code wise, it probably isn't a big change but we want to think a bit more about what does this exactly mean. I did a quick POC https://github.com/GoogleCloudPlatform/kubectl-ai/pull/333 that I will try to test it out soon and will keep you posted.

[sybadm]

@droot are you still working on this? I heavily use MCP Servers through open-webui and having tools exposed like running CIS compliance report using kube-bench via chat will be great as it it gives ability to play around the reports on LLM