Troubleshooting

The AI tool asks me to sign in again

A connection ends when it reaches the length your organization allows (90 days unless an admin chose otherwise), when it is revoked under Settings → Kubeadapt MCP, or when your access to the organization changes. Sign in again from your AI tool: in Claude Code and Codex run the login from /mcp or codex mcp login kubeadapt; in Claude, Cursor and VS Code select Connect next to Kubeadapt.

The sign-in page says Kubeadapt MCP is off

An admin turned Kubeadapt MCP off for your organization under Settings → Kubeadapt MCP → Access policy. Ask an admin to turn it back on.

The sign-in page says this AI tool isn't allowed

Your admin allows only verified AI tools, and this one isn't verified. Use a verified tool, or ask an admin to allow any AI tool.

The Kubeadapt tools don't appear

• Check that the server URL is exactly https://mcp.kubeadapt.io/mcp.
• Finish signing in; most tools list the Kubeadapt tools only after sign-in.
• Some tools read the list of tools only when they connect. Reconnect Kubeadapt, or restart the tool, to pick up tools added since.
• If only some tools appear, the connection was given only some kinds of data. Connect again and choose more.

"Cluster … is not one this connection can see"

The connection covers only the clusters you chose when connecting, within your own access. Ask about a cluster from the list get_context returns, or connect again and include the cluster.

"Too many requests"

The connection reached its rate limit of about 60 requests a minute, usually because the AI tool made many calls in a row; the message says how many seconds to wait. Wait, or ask a narrower question. A tool answer that says Kubeadapt is busy; wait a minute and try again means the same.

"The query took too long"

Long periods across many clusters can take too long to answer. Ask for a shorter period, fewer clusters, or a filter such as a namespace or team.

An answer has caveats

Kubeadapt says when a figure is incomplete instead of presenting it as a fact:

• "stopped reporting": a cluster's agent hasn't sent data recently, so its figures stop at that point.
• "lower bound": some usage had no price, so the total is a floor.
• "preliminary": the period includes today, whose data is still arriving.
• "no network data … unknown, not zero": network monitoring isn't running on that cluster, so its traffic isn't measured.
• "Kubeadapt's default rates": the cluster has no custom pricing, so its costs use default rates. Set your real prices in the cluster's settings; get_pricing shows which rates are in use.

A number differs from the Kubeadapt app

Check that both use the same window and cost mode. Answers default to the last 30 complete days in fully loaded cost, while an app page may show month to date or another range. The open_in_kubeadapt link opens the view the answer came from.

Still stuck

Contact support with the AI tool you use and the time of the failed call. The Activity view under Settings → Kubeadapt MCP shows every call and its result.