CircleCI has a short help-center article that describes a billing-grade export. Utilizing the Usage API says customers can “query your usage data for any 32-day window (or less) going back up to 13 months.” Data is refreshed daily through the previous full day, by a scheduled update that runs between 08:00 and 10:00 UTC and includes everything up to 11:59:59 UTC the day before. The export covers pipelines, workflows and jobs, credit consumption, and RAM and CPU utilization per job run. The shape is asynchronous: “Hit the API to query your data, get back a link from which you can download a CSV, then load it into a data warehouse of your choice for deep CircleCI usage analysis.”
There are no vendor figures to discount, only limits, and the limits are the story. A 32-day window and a 13-month horizon mean a full year of history is at least twelve export jobs, each returning a download link rather than a payload. The refresh window means anything queried before 10:00 UTC may not yet include yesterday. The related articles the help center lists are a map of where this surface bites: resolving a “Malformed request body” error, understanding API behavior when retrieving reports, and a Python walkthrough, which is what a job-plus-poll-plus-download pattern needs. The article also points to v3 API docs alongside v2, which is the first time a v3 surface appears anywhere in CircleCI’s record.
The catalog holds CircleCI across 18 API pages, and the Usage API is not one of them. The nearest records are the CircleCI Insights API, which the help center explicitly distinguishes from Usage in a sibling article, and the three APIs whose runs the export reports on: the CircleCI Pipeline API, the CircleCI Workflow API and the CircleCI Job API. The record’s description names the v2 REST API, the legacy v1 API, the Self-Hosted Runner API, webhooks and the Orbs Registry, and stops there. The agentic access profile maps 74 operations, 30 of them acting, none of them an export job.
The Kin Score is 72.9, exemplar band, with access clarity at 93.4 and discoverability at 81.7 against contract governance at 31.8. The Agent Readiness score is 56.1, agent-ready, with fourteen dimensions lit including the MCP server, delegated identity, protected resource metadata, dynamic client registration and the event surface. Idempotency, OpenAPI examples and dry-run mode are unlit. The measurement is strong and the gap is specific: a provider in the top band has a documented, asynchronous, credit-level usage export that the catalog has never read, and the agent that needs to answer “what did we spend on CI last quarter” cannot find it here. That is a harvest task for us before it is a score for them.