OptimNow Cloud FinOps Library
Give AI assistants practical cost guidance and searchable runbooks for cloud, AI, and data platforms.
- Source repo
- OptimNow/cloud-finops-skills
- Stars
- ★ 63
- Last updated
- today
- License
- CC-BY-SA-4.0
- Primary language
- Python
- FA score
- 65/100 · Some gaps
At a glance
- How it runs
- Works with
- Universal · cross-platformClaude CodeChatGPT · Codex · Claude.ai (Partial support)
- Cost
- Free, no paid service needed
- Setup effort
- Low · running in minutes
- You'll need
- Typical use
- A FinOps practitioner wants an assistant to reason about inference costs and capacity options for Bedrock, Azure OpenAI, or Vertex AI.
- Not a fit if
- Teams that need direct cloud-account access and automated optimization
- Teams that need live cloud and model prices inside the repository
- Source review
- 65/100 · Some gaps
What does this agent do, and when should you use it?
OptimNow’s repository packages FinOps knowledge as an agent skill and an MCP retrieval server, covering AWS, Azure, GCP, OCI, AI inference economics, Kubernetes, data platforms, allocation, anomalies, and waste detection. The skill consists of SKILL.md, references, and named-pattern playbooks; it installs as a Claude plugin or writes rules files for multiple AI tools through install.sh. The MCP server exposes six read-only tools for retrieving the same library on demand, with interactive views on hosts that support MCP Apps. It provides billing mechanics and detection playbooks, but does not read a user’s cloud account or bundle live price figures. Users run the supplied detection queries themselves and can choose the hosted MCP endpoint or the PyPI package.
Makes the content in plugins/cloud-finops/skills/cloud-finops/SKILL.md, references/, and playbooks/ available to an AI assistant. The skill path loads guidance into context; install.sh can auto-detect a target or use --tool <name>, and Claude Code users can install the plugin through its marketplace command. For MCP retrieval, a client sends requests to https://mcp.optimnow.io/mcp; the server exposes six read-only tools: list_references(), get_reference(name, section?), find_references(...), list_playbooks(), get_playbook(name), and find_playbooks(...). They return library content, faceted catalogues, or detection steps. Users then run any supplied SQL or CLI detection commands themselves. The library documents billing mechanics rather than live prices; current-price questions should use an available live pricing tool.
- A FinOps practitioner wants an assistant to reason about inference costs and capacity options for Bedrock, Azure OpenAI, or Vertex AI.
- A cloud engineer sees an unusual bill, such as NAT Gateway traffic to object storage, and needs the matching detection and remediation playbook.
- A FinOps and Finance team is evaluating prerequisites for moving shared Kubernetes costs from showback to chargeback.
- A platform team wants to retrieve runbooks for AWS, Azure, GCP, or cross-cloud waste patterns, then run the checks itself.
- A data-platform team needs billing and allocation guidance for Databricks, Fabric, or Snowflake.
- An organization is developing FinOps practices for cloud carbon reporting, region selection, or carbon-aware workloads.
How do you install or deploy this agent?
Install the skill-only plugin in Claude Code with:
/plugin marketplace add https://github.com/OptimNow/cloud-finops-skills.git
/plugin install cloud-finops@optimnowFor other supported targets, use the installer and replace <name> with a tool name supported by the script:
curl -sL https://raw.githubusercontent.com/OptimNow/cloud-finops-skills/main/install.sh | bash -s -- --tool <name>To let the script detect a target automatically:
curl -sL https://raw.githubusercontent.com/OptimNow/cloud-finops-skills/main/install.sh | bashFor hosted MCP, connect an MCP client to https://mcp.optimnow.io/mcp; the Claude Code command is:
claude mcp add --transport http cloud-finops https://mcp.optimnow.io/mcpTo install the self-hosted MCP package:
pip install cloud-finops-mcpThe README points to ./install.sh --tool mcp for client configuration snippets. It does not specify a required Python version or credentials for running the package.
How do you use this agent?
After installing the skill, ask a supported assistant a FinOps question. For account-specific questions, ask it to consult the relevant playbook, then run the returned detection query yourself. With an MCP client connected, call a read-only retrieval tool such as:
find_playbooks(scope="AWS", service="NAT Gateway")Retrieve a playbook by name with:
get_playbook(name="<playbook-name>")For reference material, use find_references(domain?, capability?, phase?, persona?, maturity?, persona_primary_only?); to retrieve a section, use get_reference(name, section?). The MCP server cannot see your cloud account: it returns guidance and detection queries for you to run. Use a live pricing tool connected to the session for current prices.
What are this agent's strengths and limitations?
- The same library is available as preloaded skill guidance or on-demand MCP retrieval.
- Coverage spans AWS, Azure, GCP, and OCI, with dedicated material on AI inference, Kubernetes, data platforms, and GreenOps.
- Six MCP tools support filtering by FinOps capability, phase, persona, and maturity, and section-level retrieval for long references.
- The repository says it is refreshed weekly and that updates are confirmed against provider documentation and reviewed.
- The tools retrieve read-only guidance; they do not connect to cloud accounts, discover resources, or execute optimizations.
- Live prices are not bundled, so current price questions require a separate live pricing tool or source.
- MCP invocation depends on the model’s tool-routing decision; the documentation reports inconsistent routing for some advisory and symptom questions.
- Hosted MCP calls go to OptimNow’s server, and arguments for unmatched lookups may be written to application logs.
How does this agent compare with similar options?
The repository presents preloaded skill guidance and on-demand MCP retrieval as two delivery modes. The skill places guidance in context before reasoning and is suited to advisory and symptom questions; MCP retrieves content through tool calls and supports discovery, faceted queries, and interactive views. They can be used together. Live pricing is outside the library and is routed to OptimNow AI Pricing Hub.
Key facts side by side with the most closely related agents.
| Agent | Source review | Form / cost | Stars | Updated | Language | Full support on |
|---|---|---|---|---|---|---|
| OptimNow Cloud FinOps Library This agent | 65 · Some gaps | Agent plugin / skillFree | ★ 63 | today | Python | Claude Code |
| Stakpak | 57 · Major gaps | CLIFree + model costs | ★ 1.8k | 3mo ago | Rust | OpenAI API · Claude API |
| OpenSandbox Runtime | 47 · Major gaps | Library / SDKFree | ★ 16k | today | Python | Codex · Claude Code |
| OpenOps | 59 · Major gaps | Self-hosted serviceFreemium | ★ 1.1k | today | TypeScript | — |
How does FollowAgents rate this agent?
Why each dimension lost points
The materials describe the skill as static text and the hosted MCP as read-only retrieval, with no access to users' cloud accounts; CI also checks read-only annotations. These support least privilege and some data-flow transparency. The README links to a privacy policy without including its full handling details here. The installer and release workflow can write files and publish externally, while the supplied evidence does not fully establish confirmation or rollback behavior. Attribution is visible, but item-level sourcing is not shown throughout the library.
The materials show layered CI checks for coverage, content sync, footers, version drift, and MCP tests, supporting internal consistency. The hosted MCP, package index, and client platforms are availability dependencies without service guarantees in the supplied evidence. User-facing runtime failure messages are also not sufficiently documented here, so those scores are limited.
The README identifies FinOps practitioners, cloud engineers, developers, and Finance/IT managers, with examples across many tasks. It clearly states that the product cannot inspect accounts or supply live prices, and discloses probabilistic MCP routing. Multiple tool installation paths and both skill and MCP delivery fit a broad set of environments. Routing still depends on the model choosing retrieval, and the README reports phrasing-sensitive results, so trigger precision is not fully established.
The repository structure, installation guide, troubleshooting, roadmap, coverage catalogues, and multi-tool commands are substantial. Known coverage gaps and unsupported MCP widgets are disclosed. Releases, automated tagging, change notes, and a maintainer are identified. The license text is clear, but metadata is NOASSERTION; the README badge and LICENSE.md corroborate the license text, while consistency of metadata across all distribution surfaces cannot be established here. Naming and examples are useful but not exhaustive.
Reference material, faceted retrieval, and pattern-specific runbooks can provide actionable detection queries and steps across cloud cost topics. The evidence for accuracy and probe performance is largely the project's own description, without independent corroboration. Retrieval costs and realized benefits across delivery modes are not quantified, limiting these scores.
Repository structure, frontmatter, automated checks, and test design provide traceable evidence for coverage and retrieval structure, and multiple files corroborate some maintenance claims. Accuracy, routing measurements, and savings outcomes are primarily asserted in the README without independent sources or complete item-level citations. Facts and inferences are partly distinguished, but not every product-effect claim is independently verifiable from the supplied material.
- The hosted MCP retrieves from a fixed knowledge library; it does not inspect cloud accounts or run detections for the user. Account-specific work requires user-provided data and execution of the suggested query.
- The README says retrieval routing can vary with phrasing. If an answer does not cite or retrieve the library content, explicitly ask for the relevant reference or playbook.
- The supplied materials do not include the full PRIVACY.md. Review that policy and the target host's data-handling terms before sharing sensitive cost or account data.
- The knowledge library excludes live prices; current pricing requires a separate connected tool or external source.
Recent releases
Latest entries from the project's GitHub Releases. Notes are quoted from upstream as written, not rewritten or tested by FollowAgents.
-
Release notes excerpt
Cloud FinOps Skill & MCP v1.42.0 The attached cloud-finops-v1.42.0.zip is the skill bundle for upload into Claude Desktop / claude.ai (Settings -> Skills -> Upload). The attached llms-full.txt is the whole library (entry point, references, playbooks) in one file, built from this tag; llms.txt links to the latest-release copy. For Claude Code users (and 11 other tools), see [INSTALLATION.md](https://github.com/OptimNow/cloud-finops-skills/blob/main/INSTALLATION.md) for the cross-tool installer and the model-agnostic response contract. What's Changed …View on GitHub ↗ -
Release notes excerpt
Cloud FinOps Skill & MCP v1.41.1 The attached cloud-finops-v1.41.1.zip is the skill bundle for upload into Claude Desktop / claude.ai (Settings -> Skills -> Upload). The attached llms-full.txt is the whole library (entry point, references, playbooks) in one file, built from this tag; llms.txt links to the latest-release copy. For Claude Code users (and 11 other tools), see [INSTALLATION.md](https://github.com/OptimNow/cloud-finops-skills/blob/main/INSTALLATION.md) for the cross-tool installer and the model-agnostic response contract. What's Changed …View on GitHub ↗ -
Release notes excerpt
Cloud FinOps Skill & MCP v1.41.0 The attached cloud-finops-v1.41.0.zip is the skill bundle for upload into Claude Desktop / claude.ai (Settings -> Skills -> Upload). The attached llms-full.txt is the whole library (entry point, references, playbooks) in one file, built from this tag; llms.txt links to the latest-release copy. For Claude Code users (and 11 other tools), see [INSTALLATION.md](https://github.com/OptimNow/cloud-finops-skills/blob/main/INSTALLATION.md) for the cross-tool installer and the model-agnostic response contract. What's Changed …View on GitHub ↗
FAQ
Will it read my cloud account or fix resources automatically?
Does using the repository require a software fee?
How do I get current cloud or model prices?
Does the hosted MCP server require an account or key?
https://mcp.optimnow.io/mcp over the network.