# Toolsets

The Kloudfuse MCP server groups its tools into **toolsets** so you can expose only the ones you need on a given connection. Selecting fewer toolsets keeps the AI assistant’s context window small, which often improves answer quality and reduces token cost.

By default the server exposes every tool—selecting toolsets is opt-in, useful when you want a smaller, task-specific surface (for example, only `apm` and `logs` for trace-driven investigations).

## In This Topic

- [Available Toolsets](https://docs.kloudfuse.com/platform/4.1.0/ai-ml/mcp-server/toolsets/#toolsets-available)

- [Selecting Toolsets](https://docs.kloudfuse.com/platform/4.1.0/ai-ml/mcp-server/toolsets/#toolsets-selecting)

- [Notes](https://docs.kloudfuse.com/platform/4.1.0/ai-ml/mcp-server/toolsets/#toolsets-notes)

## Available Toolsets

| Toolset | What it covers |
| --- | --- |
| `metrics` | Run PromQL queries against your metrics, list label names, and explore label values. |
| `logs` | Search logs, list log labels and values, and run LogQL metric queries over log data. |
| `apm` | Inspect APM services and dependencies, search and drill into traces, view trace waterfalls and span details, and surface trace errors. |
| `alerts` | Search alert rules, fetch the underlying queries, and pull alert firing history. |
| `events` | Query Kubernetes events with filtering, sorting, and facet discovery. |
| `infrastructure` | Search Kubernetes objects (pods, deployments, services, nodes, …​) and get their current status. |
| `pyroscope` | Pull continuous-profiling data: profile types, label discovery, time series, flamegraphs, and diff comparisons between profiles. |
| `rum` | Real User Monitoring — list applications, search RUM events, view error groups, fetch latency distributions and time series, and unminify stack traces. |
| `docs` | Search and retrieve official Kloudfuse documentation pages. |

## Selecting Toolsets

Append a `toolsets` query parameter to the MCP server URL with a comma-separated list of names:

```
https://<KLOUDFUSE_URL>/mcp?toolsets=apm,logs
```

The order of names does not matter. Whitespace around names is tolerated.

|     |     |
| --- | --- |
|  | Define multiple MCP server entries against the same Kloudfuse cluster — each with its own toolset — and assign them to different subagents or workflows. For example, an incident-triage subagent can use a connection scoped to `logs,metrics,alerts`, while a performance-tuning subagent uses one scoped to `apm,pyroscope`. Most MCP clients let you give each entry a separate name. |

Use this URL anywhere your MCP client accepts an MCP server endpoint — the same places shown in [Install MCP Server in Kloudfuse](https://docs.kloudfuse.com/platform/4.1.0/ai-ml/mcp-server/remote-instructions/) and the [OAuth custom connector](https://docs.kloudfuse.com/platform/4.1.0/ai-ml/mcp-server/oauth/overview/) guides. When passing the URL on a command line, quote it to prevent shell globbing or word-splitting of query characters.

## Notes

- **Applies to the remote MCP server only.** Toolset selection is read from the URL query string, so it works for the cluster-hosted server at `https://<KLOUDFUSE_URL>/mcp` and for OAuth custom connectors. The local Docker container exposes its full toolset.

- **Unknown names are ignored.** Misspelled or unsupported names are silently dropped from your selection so future toolset additions and renames stay backward compatible. Double-check the spelling against the table above if a tool you expect is not appearing.

- **Changing toolsets requires reconnecting.** Update the URL in your MCP client and restart or reconnect the connector for the new selection to take effect.