Cloud cost intelligence for AI coding agents.
demo.mp4
A collection of AI coding agent skills for Infracost users. These skills integrate Infracost workflows directly into your AI coding agent, providing cost estimation, optimization recommendations, and FinOps policy compliance for infrastructure as code.
| Skill | Description |
|---|---|
| iac-generation | Ensure cost optimization and FinOps/tagging policy compliance when writing infrastructure as code |
| scan | Analyze IaC projects to estimate cloud costs, identify savings opportunities, and flag FinOps policy violations |
| price-lookup | Look up cloud resource pricing by generating sample Terraform and running Infracost — no existing IaC required |
| fix-findings | Apply Agents FinOps findings — fix locally when the task carries the code, or have Agents draft a PR / ticket |
Works with Terraform, Terragrunt, and CloudFormation. Supports AWS, GCP, and Azure.
- A free Infracost account — sign up at infracost.io (takes under a minute; no credit card required).
Important
Make sure you've signed up for a free Infracost account before trying to use the AI skills (takes under a minute; no credit card required).
Instructions on how to install these skills in your favorite agent are available via our official AI Agent Skills docs
After installing, try any of these prompts:
Scan for costs and policy violations:
Scan this Terraform project and tell me which resources are missing required tags.
Look up pricing without any existing code:
How much does an RDS PostgreSQL db.r5.xlarge cost in us-east-1?
Optimize to a budget:
Our budget is $2,000/month. Which resources are pushing us over, and what is the quickest
way to get under budget?
Generate compliant infrastructure:
Write Terraform for an RDS PostgreSQL instance for our payments service in us-east-1.
Required tags: team=payments, env=prod, cost-center=platform. Budget: $500/month.
Fix the FinOps findings on this repo:
Walk me through my open Infracost findings and help me fix them. Open PRs via
Agents where it makes sense; do the simple ones locally.
The skills are single-sourced and written in terms of capabilities (scan, price,
policies, budgets, guardrails, and the inspect/inspect_* drill-ins). Each capability
has two thin bindings — the same Infracost engine, invoked two ways — and each agent uses
whichever fits how it works. The capability ↔ binding map is in
plugins/infracost/BINDINGS.md.
| Binding | Used by | How it's invoked |
|---|---|---|
| MCP | Claude Code, Cursor, other MCP clients | The plugin ships an MCP server (infracost mcp) that starts automatically; the agent calls typed tools and reads structured JSON — no flags, no shell pipelines, no JSON files to parse. |
| CLI | GitLab Duo, Gemini CLI, other shell-running agents | The agent runs infracost … commands directly. Commands are kept metacharacter-free (using the CLI's own --llm / --fields / --filter / --group-by flags) so Duo can offer pattern-based approval. See AGENTS.md and .gitlab/duo/chat-rules.md. |
The skill bodies hold one rule: if the Infracost MCP server is available, use its tools; otherwise run the equivalent CLI command from the matrix. So the workflow, FinOps concepts, and presentation guidance stay in one place — only the binding differs.
Note:
fix-findings(Infracost Agents) is MCP / Claude-Code-only and is not part of the CLI binding.
Under the hood both bindings connect to Infracost Cloud for live pricing data and your
organization's policies, and read your local IaC files. With MCP, auth + active organization are
resolved once at startup; with the CLI, they come from your existing infracost login (or
INFRACOST_API_KEY + INFRACOST_CLI_ORG on headless runners).
We welcome contributions! Please start by opening a thread in GitHub Discussions to discuss your idea before submitting a PR.
If you run into any issues or have feedback, please open a thread in GitHub Discussions. We'd love to hear from you!
- Website: infracost.io
- Sign up (free): dashboard.infracost.io
- Docs: infracost.io/docs/infracost_cloud/ai_skills/
- Issues / feedback: github.com/infracost/infracost/discussions