Eclipse Outwatch
Independent uptime/SLA witnessing for MCP servers: hash-chained probe receipts feed a recomputable trust record, availability recomputed only from receipts — no self-report accepted — and a badge that auto-degrades on witnessed downtime. LOCAL scope only — probes run only against localhost or fixture:// endpoints; vantage points are local-simulated probe identities, NOT geographically independent; no live external monitoring.
status: in-developmentclass: mcpTrust & ReceiptsCertification & Testing
Vendor: Cyan Eclipse Technologies
What it does
- Novel mechanism — probe receipts are hash-chained into a listing's trust record, so SLA claims are witnessed history, not vendor self-report, and a badge auto-degrades on witnessed downtime (SPEC §2): every probe appends a receipt {seq, ts, target_id, vantage, vantage_mode, scope, ok, status, latency_ms, prev_hash, hash} where hash = sha256(canonical_json(receipt_without_hash)) and prev_hash is the previous receipt's hash for that target (genesis = 64 zero hex chars). Availability and badge tier are recomputed only from receipts; no tool accepts a self-reported uptime/SLA input. Bench: the bench independently recomputed every sha256 from the SPEC canonical form — match; field/ts edits, single-receipt rehash, middle/first deletion, and reorder on store copies were all detected.
- No self-report path + badge ladder (original bench P1/P4 PASS): 14/14 canonical self-report shapes (uptime, availability, sla, sla_claim, claimed_uptime, availability_claim, self_report, self_reported_uptime, full Checkly-dashboard payload) refused SELF_REPORT_NOT_ACCEPTED on every tool, creating no target and no receipts; 12 setter/import tool names all NO_SUCH_TOOL. Exactly the 5 spec tools exist, with no setter/import/edit or badge/self-report input in any schema. Badge derived only from receipts in the window: 0 → NO_WITNESSED_HISTORY; 1–4 @100% → INSUFFICIENT_HISTORY; 5 @100% → UP; 19/20 = 0.95 → UP; 9/10 = 0.90 → DEGRADED; 4/5 = 0.80 → DEGRADED; 7/10 = 0.70 → DOWN; window honored exactly (default 20, clamps at 100). Availability math exact: localhost sequence [200,503,200,404,200,503,201] → 4/7 = 0.5714285714285714; fixture error/down → 0.0; connection-refused localhost → ok=false status=0 receipt; 0 receipts → NO_WITNESSED_HISTORY, availability null; targets isolated.
- Redirects never followed + crash-proof + structural refusal (v0.1.1 fixes for bench defects D1/D2/D3, re-bench PASS): probes use a no-redirect opener — the registered endpoint's own response (incl. 3xx, recorded with its true status) is the witnessed outcome, no same-origin/cross-origin/external hop is ever probed (re-bench 2a/2b/2c: same-origin 302→503, cross-origin 302→200, external-shaped 302, loop and chain all witness 302 only, destination hit counts unchanged 0→0). Malformed arguments can no longer kill the process: target_id type-checked in every tool, urlparse ValueError → clean validation error, dispatch + main-loop guards map unexpected exceptions to clean isError (re-bench P6: 21/21 checks, aggregating 82 fresh-process vectors, 0 crashes). Advertised schemas are enforced at dispatch (unknown keys rejected structurally) and self-report detection is normalized (case/camelCase/separators/suffixes + nested scan) — 19 variant shapes all refused, availability stayed exactly 0.5, receipt count and badge unchanged.
- Honest scope — LOCAL ONLY: probes execute only against localhost (127.0.0.1/localhost/::1 HTTP) or fixture:// endpoints; 17 external/non-local URL shapes (incl. 127.0.0.1@evil.com, hex/decimal IP, sub.localhost, file:, ftp:) all SCOPE_EXTERNAL_REFUSED; conflicting re-register TARGET_CONFLICT, identical idempotent; LOCAL + local-simulated labels verified on receipts, trust record, badge, and verify outputs; fixture seed never auto-loaded. 'Vantage points' are distinct local probe identities (local-vantage-a/b/c), labeled vantage_mode: local-simulated in every artifact — they are NOT geographically independent and are never described as such. Python 3 standard library only; MCP over stdio; exactly 5 tools. No checkout/payment code exists — the $299/mo feed license is a pricing term only.
- Disclosed LOWs, open as documented for LOCAL scope (non-blocking): D4 — the receipt hash chain is unauthenticated (plain sha256, no key/signature, as specified): anyone with write access to the store file can append a correctly-hashed forged receipt or truncate the tail undetectably (bench on store copies: forged append verified valid=true checked=7; last-receipt deletion valid=true checked=5); consumers should pin chain_head from get_trust_record externally and call verify_chain separately (get_trust_record/badge_state do not embed chain validity); a future version should HMAC/sign receipts. D5 — vantage provenance is free-form: probe_now accepts arbitrary vantage names verbatim (a name like aws-us-east-1-real-geo appears in the receipt and per-vantage trust record), mitigated not eliminated by the vantage_mode: local-simulated stamp on every receipt, and the list form of vantages is uncapped while the count form caps at 10.
Tools
register_target · probe_now · get_trust_record · verify_chain · badge_state
Platforms
- MCP over stdio
Certification
Bench verdict: CERTIFIED (LOCAL)
Certified scope: CERTIFIED (LOCAL) — Verity independent re-bench 2026-10-26, REBENCH-1 of the v0.1.1 Bench Fix D1–D3 round (repo @ b65c32e): 21/21 checks PASS, 0 FAIL (P6 aggregates 82 fresh-process vectors). D1 (HIGH — malformed arguments crashed the server process: list/dict target_id TypeError, invalid-IPv6 urlparse ValueError) FIXED: 25 original malformed cases + 57-vector fuzz sweep (wrong types on every arg of every tool, fresh process per vector) → 0 crashes, 0 tracebacks, every vector a clean JSON-RPC isError, process alive and tools/list responsive after each. D2 (HIGH — probes followed HTTP redirects, LOCAL scope bypassable by trampoline to an always-200 or external page) FIXED: redirects never followed, witnessed as-is — same-origin 302→/down(503), cross-origin 302→second-port /ok(200), external-shaped 302, redirect loop, and 302→302→503 chain all record status=302 only with destination hit counts unchanged (0→0). D3 (LOW — self-report refusal was an exact-key denylist and declared schemas were never enforced) FIXED: 19 variant shapes across all read/probe tools (uptime_pct, Uptime, UPTIME, selfReportedUptime, sla_pct, nested {meta:{uptime:…}}, unknown keys, register-time and nested-in-list forms) all refused cleanly, zero influence (availability exactly 0.5, receipt_count 2, badge unchanged). Regression spot-checks PASS: badge ladder 19×200+1×503 → availability 0.95 → UP; chain tamper on a store copy (seq-3 field edit) → valid=false, tampered_seqs=[3], pristine chain valid=true checked=6. New defects: none. Known LOWs remain open as documented for LOCAL scope, non-blocking (D4: hash chain unauthenticated — forged append / tail truncation undetectable under store-write access; pin chain_head externally, call verify_chain separately. D5: vantage names free-form and list form uncapped, mitigated by vantage_mode: local-simulated stamp). Prior verdict on v0.1.0 (VERDICT.md): CONDITIONAL (LOCAL), 72/76 checks PASS with HIGH defects D1/D2 and LOWs D3–D6 — D1/D2/D3 (and D6 bool-as-int as a side effect) closed by this re-bench. Scope = LOCAL only: probes run only against localhost or fixture:// endpoints, external probing refused; vantage points are local-simulated probe identities, NOT geographically independent; no self-report input accepted; $299/mo is a pricing term only — no checkout/payment code.
Not tested:
- External or public network probing — refused by design in v0.1.1 (SCOPE_EXTERNAL_REFUSED): no live external monitoring is performed, tested, or claimed
- Real geographic vantage independence — not in v0.1.1 by design: vantage points are local-simulated probe identities only, stamped vantage_mode: local-simulated on every receipt
- Hosted or remote serving — v0.1.1 is local stdio only; known LOWs D4 (unauthenticated chain, tail truncation undetectable under store-write access) and D5 (free-form vantage names, uncapped list form) remain documented for LOCAL scope
- Payment, checkout, or charging — none exists in v0.1.1 by design: the $299/mo feed license is commercial metadata / a stated pricing term only, and no payment or checkout code is in the server
- Dashboards, alerting, badge issuance/revocation beyond the derived badge_state, or any certification claim by the product itself — none exists in v0.1.1; certification is Verity's bench verdict only
Live state note: Bench evidence: argus-batch-20261026/bench/outwatch/VERDICT.md + REBENCH-1.md (Verity, independent bench over stdio JSON-RPC only, never importing builder code, all hashes recomputed independently from SPEC §2; builder suite 29/29). This bench issued no test-bench/results scorecard; scorecard_ref is null and no scorecard is claimed or hosted for this product.
Pricing
PRICED — $299/mo feed license, via the Eclipse Bazaar bench-order path only. No checkout, payment, or buy-now link exists for this product; there is no payment or checkout code in this package (the price is a stated term only). Repository is PRIVATE (priced-product policy).
Install / verify
No install command is published for this product in the catalog or its registry record.
Official MCP Registry name: io.github.CYAN-ECLIPSE-TECHNOLOGIES-LLC/eclipse-outwatch-mcp
Registry listing: https://registry.modelcontextprotocol.io/v0.1/servers/io.github.CYAN-ECLIPSE-TECHNOLOGIES-LLC%2Feclipse-outwatch-mcp/versions/0.1.1
Repository: https://github.com/CYAN-ECLIPSE-TECHNOLOGIES-LLC/eclipse-outwatch-mcp
Verify any MCP server in your client configs against this catalog with bazaar check (Eclipse Bazaar MCP / CLI — see llms.txt).