Read-only cost model of an existing multi-account estate

This recipe builds a read-only, all-environments cost model of infrastructure that already exists — several hundred resources spread across a handful of accounts/environments, on one canvas — starting from terraform show -json state you already have. No plan, no refresh, and no credentials beyond a local file (or S3) read.

It contrasts with the other two ingestion paths:

The total is deliberately a rough, us-east-1 heuristic — and a floor, not a bill. See Estimated Cost for what the heuristic covers.

1. Export state per account/environment

For each account/environment, capture the state Terraform already knows about — no apply, no plan:

terraform show -json > prod.tfstate.json      # in the prod workspace
terraform show -json > staging.tfstate.json   # in the staging workspace

(If your state lives in S3, aws s3 cp s3://…/terraform.tfstate . first — that S3 read is the only credential this path needs.)

2. Cost a single file

The fastest loop is the offline CLI. It imports the state, totals it, rolls the spend up by category, and — prominently — flags the total as a FLOOR with the billable types it couldn’t price:

npm run strata -- cost ./prod.tfstate.json
# 214 resource(s): $12.4k/mo  (FLOOR — lower bound)
#   database        $6.1k/mo  (18)
#   analytics       $3.5k/mo  (5)
#   compute         $1.9k/mo  (61)
#   …
#   7 unpriced resource(s) excluded from the total.
#
# ⚠ FLOOR: the total is a lower bound — 3 billable type(s) unmapped:
#   app-runner, kendra, quicksight

Add --json for machine output (CI/agents). Tune the realism knobs when you want something closer to your actual bill:

npm run strata -- cost ./prod.tfstate.json \
  --region-multiplier 1.08 \   # e.g. a pricier region than us-east-1
  --hours 730 \                # billable hours per month
  --discount 30                # Savings Plan / RI discount, %

3. Combine every account onto one canvas

To model the whole estate together, import each state file and merge them. Merging namespaces every id per source (so ids from different accounts can’t collide) and lays the sources out side-by-side — never stacked at the origin.

From an agent over MCP:

// Import each large state file server-side and keep a handle (the multi-MB
// document never round-trips through the agent's context):
import_iac({ "path": "./prod.tfstate.json" })      // → { graphId: "graph-1", … }
import_iac({ "path": "./staging.tfstate.json" })   // → { graphId: "graph-2", … }
 
// Combine (returns the merged graph inline), then cost the whole estate:
merge_graphs({ "graphs": [
  { "name": "prod",    "graph": /* graph-1 */ },
  { "name": "staging", "graph": /* graph-2 */ }
], "wrapAs": "container" })   // → { graph, sources, resourceCount, rootCount }
 
estimate_cost({ "graph": /* merged graph */, "assumptions": { "discountPct": 30 } })

wrapAs: "container" nests each source under a labeled account block; omit it to let the layout grid arrange the sources on its own.

4. Render a diagram headlessly

Save the combined graph as JSON and render a standalone SVG — the headless equivalent of the in-app image export, no browser required:

npm run strata -- render ./estate.json -o ./estate.svg
# Wrote ./estate.svg — 512 node(s), 84213 bytes.

Why the total is a floor

A resource whose service Strata can price contributes to the total. A resource whose service is billable but not yet priced is excluded from the total and listed under the floor flag — so a redundant serverless minimum repeated across N accounts can never masquerade as $0. Free control-plane services (IAM, ACM, CloudFormation, Organizations, …) are classified separately and do not trip the flag. When the floor flag is absent, every unpriced resource was a known-free service and the total is complete for what’s modeled.