# Cloud FinOps [Health: Active]

**Category:** 💻 Developer Tools  
**Repository:** https://github.com/OptimNow/cloud-finops-skills  
**GitHub Stars:** 57  
**Views:** 1  
**Installs:** 0  
**Upvotes:** 0  
**Directory Page:** https://allmcps.com/mcp/cloud-finops

## Description
Queryable Cloud FinOps reference library and named-pattern waste playbooks by OptimNow.

## Claude Desktop Quick Installation
Install path detected from listing signals. Uses `uvx` (confidence: high):

```json
"mcpServers": {
  "cloud-finops": {
    "command": "uvx",
    "args": ["cloud-finops-mcp"]
  }
}
```

## Documentation

## What Cloud FinOps does

Cloud FinOps MCP server exposes a curated knowledge library for cloud cost management through the Model Context Protocol. The content covers AWS, Azure, Google Cloud, Oracle Cloud Infrastructure, AI cost management and inference economics, Kubernetes, data platforms, allocation, chargeback, anomaly management, and named-pattern waste detection playbooks.

The material is intended for FinOps practitioners, cloud engineers, developers building internal FinOps agents, and teams evaluating AI-assisted cost analysis. It provides reference guidance and runbooks rather than direct access to billing systems or cloud resources. For questions about your own reservations, usage, or waste, the relevant playbook supplies a detection query or procedure that you run against your data.

## How it works

The Cloud FinOps MCP server presents the library as read-only retrieval tools that an MCP-compatible client can call when it decides more context is needed. Retrieval is on demand, so the model may not call a tool for every advisory or symptom-based question. Asking the agent to check the playbook library or show the relevant runbook can make the intended retrieval explicit.

A hosted endpoint is available at `https://mcp.optimnow.io/mcp`. The project also publishes an installable `cloud-finops-mcp` Python package. Hosts that support MCP Apps may render interactive widgets, while the underlying content remains a reference and retrieval surface.

## Setup and configuration

For a hosted connection, add the endpoint as an HTTP MCP server. Claude Code can use:

```bash
claude mcp add --transport http cloud-finops https://mcp.optimnow.io/mcp
```

The URL must use `/mcp` without a trailing slash. Claude.ai and Claude Desktop can add it through Settings, Connectors, and Add custom connector. For a local package installation, run:

```bash
pip install cloud-finops-mcp
```

Then add the package to the MCP client configuration. The repository also provides an installer with an MCP target and installation paths for several supported agent tools. Full installer options and troubleshooting are documented in the project repository.

## Tools and capabilities

The Cloud FinOps MCP server provides six retrieval tools for querying the library. The README describes the server as supporting:

- FinOps reference knowledge across major cloud providers
- AI and inference cost analysis
- Kubernetes and data-platform cost topics
- Allocation and chargeback guidance
- Anomaly-management material
- Named-pattern waste playbooks
- Faceted queries over library metadata
- Runbook retrieval for investigation and detection workflows

The server is read-only. It does not inspect cloud accounts, change resources, execute billing queries in your environment, or return account-specific findings without user-supplied data.

## Limitations and notes

The Cloud FinOps MCP server does not replace account-level analysis. It cannot determine which of your reservations are expiring, identify your actual idle resources, or validate current spend by itself. Use its playbooks to obtain the relevant query or investigation method, then run that method against your own exports and systems.

Tool selection is probabilistic because the client decides whether to call the server. Lookup and discovery questions are more likely to retrieve material than broadly phrased advisory questions. The repository reports that equivalent questions can produce different grounding, so explicitly requesting the library or a named runbook is useful when no tool call is visible.

The project offers both a skill-based delivery model and the MCP server. A skill places the files into an agent's context, while MCP retrieval fetches material as needed; these are different delivery mechanisms for the same library.

_Full upstream README: https://allmcps.com/mcp/cloud-finops/readme_