Infracost Price Lookup

Price Lookup

Look up cloud resource pricing without needing existing infrastructure code. Supports any resource type that Terraform and Infracost support across AWS, GCP, and Azure.

You synthesize a small Terraform snippet for what the user is asking about and price it. The result is a per-resource breakdown plus a headline summary.

Invoking capabilities

This skill uses the price capability (and optionally the inspect capabilities to reproject the result). Each has two bindings — a typed MCP tool call and an infracost … CLI command — listed in BINDINGS.md.

Rule: if the Infracost MCP server is available, use its tools; otherwise run the equivalent CLI command from BINDINGS.md. With MCP, the snippet is passed as a string to the price tool. With the CLI, the snippet is fed to infracost price on stdin via a heredoc (this is the one command in the skill set that needs a shell metacharacter — see BINDINGS.md).

Before pricing, satisfy the three preflight capabilities (see BINDINGS.md): the CLI is present, the user is authenticated, and an organization is selected. With MCP these are resolved on the first tool call (failures come back as readable errors, not a dropped connection); with the CLI, check them inline. Never run infracost auth login — it is interactive.

Workflow

1. Write minimal Terraform

Synthesize a small Terraform snippet for the resource(s) the user is asking about. Rules:

• Minimal config only — no backends, no variable files, no outputs, no data sources. Just a provider block and the resource(s).
• Include attributes that affect pricing — instance type, storage size/type, engine version, throughput, IOPS, etc. These are the knobs that change the price, so they must be present.
• Use the user's requested configuration — if they ask for "m5.xlarge with 100GB gp3", write exactly that. If they ask generically ("how much does an RDS instance cost?"), pick reasonable defaults and clearly state what you chose.
• Set the region — use the region the user asks for, or default to us-east-1 and mention it.
• Multiple resources are fine — if the user asks about several resource types, put them all in the same snippet.
• Use realistic names — name resources descriptively (e.g. aws_instance.web_server, not aws_instance.example) so the output is easier to read.

2. Price the snippet

Use the price capability with the Terraform snippet (see BINDINGS.md for the exact MCP tool / CLI invocation):

provider "aws" {
  region = "us-east-1"
}

resource "aws_instance" "web_server" {
  ami           = "ami-0c55b159cbfafe1f0"
  instance_type = "m5.xlarge"

  root_block_device {
    volume_size = 100
    volume_type = "gp3"
  }
}

For non-USD currencies pass a currency (ISO 4217 code — EUR, GBP, JPY, CAD, AUD, …). If the user doesn't specify a currency, the org-configured default is used (typically USD).

3. Read the result

A price returns:

• currency — the ISO code in use.
• summary — headline counts (resource counts, monthly cost, etc.) — the same shape a scan returns.
• resources[] — one row per top-level resource you sent in, with:
  • name, type, is_supported, is_free
  • total_monthly_cost — pre-summed across cost components AND any nested subresources (e.g. an EKS NodeGroup's LaunchTemplate + EBS volumes are folded into the parent's total).
  • cost_components[] — per-component breakdown (vCPU, storage, IOPS, …) with unit, price, quantity, and the monthly costs.
  • tags, supports_tags, supports_default_tags, metadata (filename + line in the synthesized Terraform).

Unlike a scan, a price always includes the per-resource list — you've just shipped the IaC, so the cost of returning it is fixed and small.

4. Optional: drill in with the inspect capabilities

The price result is cached just like a scan result. If you want to project the data differently (e.g. group by type, top N by cost), use any of the inspect capabilities — they all read the latest cached result by default:

• summary — same summary block rebuilt through the inspect filter pipeline (project to specific fields or scope by project / provider).
• top N savings — top N FinOps savings opportunities on the priced resources (if any).
• aggregate resources by dimension (e.g. by type, provider) — reproject the per-resource list.

These rarely add value over reading resources[] directly when the snippet is small, but they're handy for "what would the top 5 savings be on this?" follow-ups.

Presenting Results

Lead with the monthly cost — that's what the user came for.

• Total monthly cost first, then the per-component breakdown.
• Call out usage-based costs. Some line items (data transfer, requests, etc.) depend on actual usage and use typical-default assumptions. Be explicit about what the estimate assumed.
• State the region. Pricing varies by region — always say which one you used.
• Compare side-by-side when asked. For "m5.xlarge vs m5.2xlarge" questions, price once with both resources in the snippet and present a comparison table.
• Surface FinOps recommendations. If the result shows applicable FinOps policies (e.g., "use GP3 instead of GP2", "consider Graviton"), highlight them with the potential savings.

Example presentation

AWS RDS MySQL — db.r5.xlarge, 100GB gp3 Region: us-east-1

Component	Monthly Cost
Instance (on-demand)	$365.00
Storage (100GB gp3)	$11.50
Total	$376.50

Usage-based costs (estimated):

• I/O requests: ~$X/mo based on typical usage

Savings opportunity: Consider Graviton (db.r6g.xlarge) for 20% savings ($876/yr).

Important Guidelines

• Do not commit any generated Terraform — the snippet you build is throwaway, never written to the user's workspace.
• Do not modify the CLI source code unless the user explicitly asks for it — this skill is for using Infracost.
• If a price reports an authentication error or "no organization selected", relay the actionable message back to the user — don't retry blindly. Auth + org are resolved on the first tool call, so the message tells the user what to do: run infracost auth login in their own terminal (interactive — never run it yourself), or infracost org switch <slug> for one of the slugs the error lists. The running MCP server picks up a saved org selection on its next call.
• If you're unsure of the Terraform resource name for what the user is asking about, look it up rather than guessing — an incorrect resource type will produce no pricing data.