Averth MCP is a read-only analysis service. You submit a sanitized batch of agent run records (cost and timing metadata only) and it returns computed economics: fully-loaded cost per accepted outcome, expensive-tail share, and offline policy replays. You can also submit human-written acceptance checks and recorded runs of coding-agent work (tool verify_evidence), and it returns an evidence ledger (VERIFIED, FAILED, NOT OBSERVED, EXTERNAL-UNVERIFIABLE, or NOT APPLICABLE per check). It never enforces anything, never executes a command, never reads a repo, never repairs code, and never guards a merge. It requires no account.
Only the run ledger you submit in a request: per-run cost figures, token counts, retry counts, outcome flags, and timing metadata. For evidence verification, also the human-written checks, recorded runs (command, timestamps, exit code), and artifact snapshot digests you submit. Please do not submit raw customer content — no prompts, completions, tool payloads, or personal data. The service is designed for metadata only. Warning: averth ledgers export caller free text verbatim, including case_id, tool names, and retry reasons, and that free text can contain customer data, so redact sensitive values before submitting; this service makes no sanitization guarantee on free-text fields.
Submitted ledgers are processed in memory and never written to disk, never logged, and never retained after the response is returned. There are no accounts, no cookies, no analytics, and no tracking.
We do not sell, share, or disclose submitted data to any third party. The only data that leaves our system is the analysis result returned to you in the same request.
Questions about this policy: [email protected].