Cyan Eclipse Covenant MCP — Cyan Eclipse Bazaar

Vendor: Cyan Eclipse Technologies · Platforms: MCP over stdio

Certification scope

CERTIFIED(local/simulator) — Eclipse Bazaar Test Bench 2026-10-25, standing card eclipse-covenant-mcp-2026-10-25-4.json (scoped re-bench -4): 40 pass, 0 na of 40 checks (40/40 scoped lines: 37 independent harness lines + builder-suite line + surface line + workspace-integrity line). CEILING — read exactly: P0 is SYNTHETIC ONLY. No live outreach of any kind is certified or claimed; a counsel HARD GATE stands before any P1 live outreach, and it has not been lifted. Defect history, all retained append-only: base card eclipse-covenant-mcp-2026-10-25.json FAILED (28/35 — D1 CRITICAL revocation-wipe, D2 HIGH provenance downgrade); -2.json PARTIAL (44/45 — D1–D7 closed, D8 open); -3.json PARTIAL (27/29 — D8 closed on its shapes, D8b open at two entry points); -4.json closes D8b at all three entry points including the Gate state-tamper path (VENDOR_POSTURE_UNKNOWN, 0 sends, refusals sealed) with sweep-hardening verified fail-closed and no over-refusal; B5 posture exhaustion and rule-pack rigidity pass. BACKLOG: LOW notes from cards -2/-3 (B1 refusal sealing / stale pack at Gate only / clarification 5-minute window) remain open, LOW severity, recorded as not gating for P0 local/simulator certification. N-CHECK: NOVELTY_PROVEN (stands from bench 1; the D8b fix hardened the proven mechanism, changed no surface).

What it does

Every lead carries a Herald-signed lineage record — CERTIFIED_CAPTURE / SELF_CAPTURED / VENDOR_ASSERTED / LEGACY_UNVERIFIED, never blended — and outreach executes only through a send-time Gate inside the contact tools: mandate × per-channel consent scope (voice consent is not text consent) × named-seller coverage × no open revocation × fresh DNC scrub at send × fair-housing guards
Consent and revocation rules are versioned DATA packs (PEWC + April 2025 revocation rules, 10-business-day honor), never code constants — missing or stale packs fail closed, and unknown vendor-posture values refuse by name at pack load, mandate create, and the Gate (state-tamper included)
Every send and every refusal seals into a Blackbox-form evidence pack addressed to the broker of record, verifiable offline via consent_record_verify
Copy Screen + Criteria Screen run a closed findings vocabulary and never silently rewrite; NO data/scraping layer exists — Covenant never holds or redistributes MLS data, scrapes portals, or resells/transfers/pools leads

Tools

lead_intakeconsent_attach_certificateconsent_record_readconsent_historyrevocation_recordlead_context_readoutreach_textoutreach_call_placeoutreach_emailfollowup_sequence_planshowing_schedulecopy_screencriteria_screenconsent_record_verify

Problems it addresses

My agents work real-estate leads, and I need consent enforced at send time — per channel, per seller, revocation-aware — not a policy document the agent is trusted to remember
When a lead says 'stop', I need that revocation to survive re-imports, duplicates, and tampered state — and proof addressed to my broker that it did
I need vendor-asserted leads held to a narrower posture than leads I captured myself, with the classes never blended in reporting

Not tested

Live outreach of any kind — P0 is synthetic only; counsel HARD GATE stands before any P1 live outreach
Real DNC registry access, a real CRM (Follow Up Boss) connection, or real TrustedForm/Jornaya verification — all fixture-shaped at P0
The backlog LOW items from cards -2/-3 (B1 refusal sealing / stale pack at Gate only / clarification window) — not re-benched in the -4 scope; status stands as recorded there

Bench scorecard: test-bench/results/eclipse-covenant-mcp-2026-10-25-4.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 →