Vendor: Cyan Eclipse Technologies · Platforms: MCP over stdio
Certification scope
CERTIFIED(local) — Eclipse Bazaar Test Bench 2026-10-25, 18 pass, 1 na of 19 checks (U1 auth: stdio process boundary). Run by W-003 (Verity), suites/umbra.py. Run 1 PARTIAL card retained append-only (bench-suite assertion defect, disclosed on the card; no product code touched); run 2 is the standing card; zero product defects. N-CHECK: NOVELTY_PROVEN — a closed-vocabulary witness: signed findings with reproducible evidence packs and rebuttal linkage, not diffs. Non-failing observation on the card: punycode-shaped domains are not IDNA-decoded by the skeleton engine; IDNA-decode is recommended before G-LIVE.
What it does
- Cross-registry lookalike, typosquat, and silent-endpoint-change watch for agent identities: imports registry snapshots (content-addressed; listings treated as untrusted data), diffs consecutive snapshots, and computes confusable-skeleton similarity against operator watchlists
- The skeleton is case-aware (capital I confuses with l; lowercase i does not) and every similarity record carries its algorithm, both claim refs, and the exact confused characters, so anyone can recompute it
- Findings issue ONLY in a closed state vocabulary (CONSISTENT / CHANGED / DIVERGED / UNREACHABLE / EXPIRED / REVOKED_BY_ISSUER / DISPUTED / UNRESOLVED), each backed by a complete reproducible evidence pack — an incomplete pack is refused by name (EVIDENCE_PACK_INCOMPLETE)
- Findings are signed under a C1 findings key structurally distinct from the Herald root identity key; wrong-key and tampered-body verification fail
- A subject's rebuttal is stored as subject-presented data and linked; the original finding stays byte-unchanged and a DISPUTED version supersedes it; accusation vocabulary is banned at render by the output lint
- No enforcement capability at any tier: no delist, block, takedown, or report-to-registry verb exists (T3 is EMPTY) and the denylist is refused at the transport. v0 is snapshot-import only; live crawling is gated per registry (G-LIVE) and NOT built
Tools
snapshot_importwatchlist_setdiff_runsimilarity_scanfinding_issuefinding_getfinding_listfinding_verifyrebuttal_filereceipt_listreceipt_verify
Problems it addresses
- Someone registered a lookalike of my agent's name on another registry and agents are hiring the copy
- A counterparty silently changed its endpoint — I need the diff with both snapshot hashes, not a rumor
- I need a witness that states findings as evidence I can recompute, never as accusations
Not tested
- Live registry observation — v0 imports snapshots only; live crawling is G-LIVE per registry (spec 5.6) and is NOT built
- Distribution of findings (T2 watch bulletins) — operator-mediated until G-REG + G-COUNSEL per spec chapter 7
- Real registry snapshots — v0 benches against fixture registries shaped like real exports; no real registry was contacted
- Findings-key rotation/revocation distribution via Eclipse Revoke — Revoke is a later C1 chapter; rotation is rehearsed at the key-file level only
Bench scorecard: test-bench/results/eclipse-umbra-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 →