Natural language to production queries.
Type a question in plain English. The MCP Server translates it into FuseQL, PromQL, or LogQL, executes against your data lake, and returns correlated results. No query language required.
Kloudfuse MCP Server gives AI agents governed, natural-language access to your production metrics, traces, logs, and dependencies. Deployed in your VPC. Authenticated against your platform's access controls. Audit-logged every time.
kloudfuse / mcp / checkout-svc · live · claude desktop you ▸ Why is checkout-service slow? ✓ identify entity · checkout-service 40ms ✓ fetch metrics · p99 latency, error rate 120ms ✓ retrieve traces · 1,204 spans 310ms ▸ search logs · error fingerprints … root cause candidate p99 latency rose 3.2× after deploy v2.4.1 — pod checkout-7f9 on saturated node node-7.
Type a question in plain English. The MCP Server translates it into FuseQL, PromQL, or LogQL, executes against your data lake, and returns correlated results. No query language required.
Built-in query guardrails keep agent-generated queries bounded, and every AI-generated query is validated before it runs. The AI operates with the requesting user's permissions, never higher.
OAuth-based one-click install for Claude Desktop, Claude Code, OpenAI Codex, Gemini CLI, and ChatGPT. Authenticate once, connect instantly — one service account token, three parameters, minutes to deploy.
A platform deep dive: structured investigations, three layers of dependencies, hard safety rails, AI-assisted root cause, and the toolsets underneath.
Ask "Why is checkout-service slow?" and the MCP Server runs a structured sequence — entity identification, metric fetch, trace retrieval, log search, dependency check — and returns a correlated response with root cause candidates, not raw query output.
Kloudfuse maps dependencies at three distinct layers derived from live production data. When an AI agent asks "what changed?", it traces impact across all three in a single query — service to workload to underlying node.
Every MCP request authenticates against your platform using OAuth. The AI operates with the requesting user's permissions, never higher. Every invocation is audit-logged with full context.
| User | Prompt | Query | Status |
|---|---|---|---|
| a.rao | checkout p99 last 1h | PromQL | ✓ ok |
| j.kim | errors by endpoint | FuseQL | ✓ ok |
| m.singh | all logs, all time | LogQL | ✗ blocked |
| l.chen | cart latency vs deploy | PromQL | ✓ ok |
| d.osei | node-7 saturation | FuseQL | ✓ ok |
Kloudfuse MCP Server exposes specialized tools spanning Prometheus metric queries, APM trace analysis, Kubernetes entity lookups, log search and fingerprinting, alerts, events, profiling, and RUM session analysis. The AI client selects the correct toolset based on the question — no manual selection needed.
| Toolset | Query surface |
|---|---|
| Metrics | PromQL range queries · p50/p99 · rate — auto-selected for "p99 latency" questions |
| APM traces | FuseQL span search and trace analysis |
| Kubernetes | Entity lookups across pods, nodes, namespaces |
| Logs | LogQL search with fingerprinting |
| Alerts | Alert management and status |
| Events | K8s and platform event lookup |
| Profiling | Flame graph and profile queries |
| RUM | Session and Core Web Vitals analysis |
Connect your AI client, authenticate with a service-account token, and ask your first natural-language question against production data. Guides for Claude Desktop, Codex, and Gemini CLI included.
Read the guideQuery safety, access controls, audit logging, and what it takes to connect AI agents to production observability data without compromising governance.
Read the blogVPC deployment, query safety controls, audit logging, OAuth integration, and AI client support. A side-by-side breakdown for teams evaluating both.
See the comparisonThirty minutes on your telemetry. The cause, before the call ends.