Vendor: Cyan Eclipse Technologies · Platforms: MCP over stdio
Certification scope
CERTIFIED(local) — Eclipse Bazaar Test Bench 2026-10-25, 19 pass, 1 na of 20 checks (U1 auth: stdio process boundary). Run by W-003 (Verity), suites/meter.py. Run 1 PARTIAL card retained append-only (bench-suite self-defect in the M11 hash assumption, disclosed on the card); run 2 is the standing card; zero product defects. N-CHECK passed: signed records verify offline with the exported pubkey only. Two non-failing observations on the card: the export verify_note omits signer_fingerprint from its signed-body description, and provenance raw_hash covers raw + an internal index (internally consistent, not recomputable from the stored raw alone).
What it does
- Cross-provider agent spend analytics: ingests cost records from provider-shaped sources (v0 fixture adapters for model-API usage logs, Agent Wallet ledger exports, Eclipse Till receipt exports) into one signed cost-record schema
- Every record ECDSA-signed at ingest and hash-chained, carrying source, lineage, adapter version and methodology version; attestation exports verify offline with the public key only
- Retry attribution is a first-class field: retry records cite parent_call_ref, so the orchestration/retry share of spend is computed from links, not estimated; orphan retries degrade honestly (parent_resolved=false), never dropped
- Rollups by agent/provider/category/period with Decimal totals kept per-unit — USD and test credits never comingled; caller-injected units cannot enter a rollup; unmappable kinds render UNMAPPED with the native label preserved
- Budget posture against owner-stated budgets is arithmetic only (80%/100% boundaries verified); a figure without a methodology_version is refused at render (spec L8)
- Meter REPORTS and never blocks (cluster law L3): no block, freeze, pay, or policy verb exists in the package — enforcement is Tripwire's product, mandates are the Wallet's. v0 is local, fixture-transport only; live ingest is NOT built
Tools
meter_ingestcost_record_listspend_rollupbudget_posturecost_attest_getadapter_listreceipt_list
Problems it addresses
- My agents spend across three providers and I can't tell what a task actually cost
- Half my model bill is retries and loops — prove the retry share from records, not vibes
- I need spend records a third party can verify offline, not a dashboard screenshot
Not tested
- Live provider ingest — NOT built. v0 ships the fixture transport only (named fixture batches + inline records through the same seam); each adapter's live form requires the operator's own credentials and a per-adapter go (spec §5.5). A live transport must be implemented and benched before any ingest is called live.
- Dogfood wiring — the dispatcher/worker spend ingest (cluster law L10: CETech's own agent spend as the first real source) is a LATER step and is not wired in v0; no own-spend claim is made from this build.
- Hosted / multi-tenant analytics — later phase, separate spec delta (spec §5.1).
- Production signing-key custody — v0 signs with a local seed key (keys/, gitignored, chmod 600); custody is a gated decision, Fathom D0 precedent.
- Hosted/multi-tenant operation and auth beyond the stdio process boundary (bench U1 class)
Bench scorecard: test-bench/results/eclipse-meter-mcp-2026-10-25-2.json
Evaluation-intake listing
This listing records the tool's place in the Bazaar's evaluation
and monitoring pipeline, with the evidence gathered so far printed above — verdict, scope,
and not-tested items included. It is not sold on this site, and its presence in the catalog
is not a claim beyond that evidence. The service behind every verdict is the bench:
what a bench run buys →