Consumption spend is the hardest AI line to govern
- What are we actually spending on Anthropic models this month, and why did it move?
- Which models and workspaces drive the majority of that cost?
- How much of it is Claude Code, and how many developers does that represent?
- Does this consumption overlap with a seat license we also pay for?
- What is our total Anthropic relationship worth across both products?
Because usage is variable, consumption spend tends to be reviewed only when an invoice surprises someone. Portfolio reporting makes it a standing line rather than an incident.
What Anthropic provides — and where the boundary sits
The Anthropic Console reports usage and cost for your organization, with model and workspace breakdowns and, where enabled, Claude Code analytics. It is the authoritative source for what the platform charged you, and Midgentic reads it directly.
What it cannot show is the same money in context: your seat licenses, other vendors' consumption, or one normalized view of AI investment across the company. That is where a portfolio layer is required.
What Midgentic actually ingests from Anthropic
- Metered API spend
- Cost reporting from your organization, normalized for currency and minor units so rollups stay exact.
- Model-level usage
- Consumption by model, so cost movement can be traced to a specific change in workload.
- Workspace attribution
- Usage split by workspace, so consumption maps to teams and initiatives rather than one undifferentiated total.
- Claude Code analytics
- Developer-side activity where your organization has Claude Code enabled.
- Usage trends
- Historical series rather than a current-month snapshot, so anomalies are visible.
- Value assumptions (entered by you)
- The assumptions behind estimated value, kept separate from measured outcomes.
No prompt, completion, or code content is collected.
The questions this data answers
- Consumption spend
- Real metered cost per period, in your reporting currency.
- Cost drivers
- Which models and workspaces account for the change, not just the total.
- Developer AI usage
- Claude Code activity as a distinguishable part of the consumption picture.
- Vendor consolidation
- Anthropic seat licensing and metered usage in one vendor total, counted once.
- Portfolio share
- How much of total AI investment is variable consumption versus fixed licensing.
Metered AI belongs in the same report as licensed AI
Treating consumption and licensing as separate worlds is how AI budgets lose coherence. Midgentic normalizes both into one model so spend intelligence and ROI cover the whole portfolio, including Claude Enterprise under the same vendor.
Common questions
What Anthropic API data does Midgentic collect?
Usage and cost reporting from your Anthropic Console organization: metered spend, token and request usage by model, workspace-level attribution, and Claude Code analytics where your organization has it enabled.
Does Anthropic API spend sync automatically?
Yes. Unlike seat-licensed products, the Anthropic platform reports metered cost, so spend is synced rather than entered. Amounts are normalized so currency and minor-unit handling stay correct in portfolio rollups.
Does Midgentic see our prompts or completions?
No. The connector reads usage and cost reporting endpoints. Request and response content is never accessed.
How does this relate to Claude Enterprise?
They are separate providers with separate cost models — metered consumption here, seat licensing there. Midgentic reports each individually and rolls both into an Anthropic vendor total without counting the same money twice.
Is the Anthropic API connector generally available?
It is implemented and in beta, enabled per workspace during onboarding while it is validated with early customers.