Skip to main content
Glama
Cyan Eclipse License Server

Cyan Eclipse License Server

Official
README.md
# Cyan Eclipse License Server

**Powered by Cyan Eclipse Technologies** — https://cetech.ai

**Provenance. Governance. Proof.**

Cyan Eclipse License Server is the firm's licensing system of record: an
MCP-native core that registers products and versioned tiers, issues signed
`.cetlic` licence artefacts (Ed25519 over canonical JSON) under a
propose → mandate-confirm flow, records every lifecycle operation in a
per-tenant append-only hash-chained issuance journal, and — under the
owner-ruled activation model (B) — runs online activation with real-time
seat control: a seat is a live activation slot, check-ins carry an
activation token and artefact hash only (no usage telemetry), revocation
takes effect at the next check-in, and a bounded 72-hour offline window
separates a transient outage from suspension. A licence revoked under
the firm's refund/chargeback standard (Standard C) can be reinstated —
`license_reinstate_propose` → `license_reinstate`, propose →
mandate-confirm — returning it to ACTIVE on the same licence id when
the payment is reversed in the firm's favour, with the reinstatement
journal-linked to the revocation it reverses.

Status: v1 core built 2026-11-12 and benched **PASS-WITH-FINDINGS at
tenant-#1 fixture scope** (Verity bench, 2026-11-12) — that verdict
covers the fixture scope; it is not a production certification. Cyan
Eclipse runs this server for its own applications. This repository is
the open MCP server for **self-hosting under MIT**. External hosted
tenancy is v2 scope and is **not open today** — there is no public
sign-up; hosted licensing is arranged by quote.

## Layout

- `src/license_server/` — the core (store, journal, artefact, keys),
  the MCP stdio server, the thin HTTP face, and the tenant-#1 seed.
- `client/` — the client library: offline artefact **authenticity**
  verification (SPEC S2.3 steps 1–6) plus an activation client and the
  model-(B) client state logic. Honest limit, stated in the package
  itself: offline authenticity ≠ entitlement to run — under (B),
  entitlement additionally requires a live activation in good standing.
- `tests/` — the unittest battery (journal chain, artefact mutation,
  mandate gate, seat cap, revocation propagation, stale reclaim,
  cross-tenant isolation, HTTP↔core parity, client offline states).

## Use

```
PYTHONPATH=src python3 -m license_server --seed     # seed tenant #1 (Cyan Eclipse)
PYTHONPATH=src python3 -m license_server            # MCP server on stdio
PYTHONPATH=src python3 -m unittest discover -s tests -v
```

Any MCP-capable host can drive it over stdio — point the host's MCP
configuration at the server command:

```json
{
  "mcpServers": {
    "cyan-eclipse-license-server": {
      "command": "python3",
      "args": ["-m", "license_server"],
      "env": { "PYTHONPATH": "src" }
    }
  }
}
```

Caller context rides in the tool arguments (`_tenant_id` /
`_principal_id`); authorisation — principal class, mandate scope,
tier floors — is enforced inside the core, and every tool's contract
is documented in its own description, input schema, and output
schema on the `tools/list` surface.

Runtime state (SQLite DB + signing keys) lives OUTSIDE the repo tree in
`cyan-eclipse-license-server-runtime/` beside the build home: the signing
private key is a locked 0600 file in a 0700 directory, never in git,
never in the DB (the key registry holds public keys + kids only).

## Policy defaults (as coded, published by `policy_show`)

heartbeat 4 h · offline window 72 h · reclaim window 30 d · expiry grace 14 d.

## Standing notes

- Tenant #1 is seeded with products `parity` / `loom` / `plumb` and their
  SPEC tiers as fixture-grade wiring readiness only — no integration into
  the actual apps in v1 core.
- Per the firm's new-app branding default, branding is applied at
  creation (see `BRANDING.md`); this product's icon is **not** invented
  here: icon election is the owner's act and is pending, gated by Icongate.
- Known open gap, on the record (SPEC S12.5): model (B) cannot activate
  in disconnected / air-gapped environments; v1 contains no offline
  activation path and no fallback. Owner ruling owed.

A Cyan Eclipse Technologies product. The "Powered by Cyan Eclipse Technologies" attribution travels with it.

## Web admin (SPEC S1.7)

`admin/` in this repo is the minimal web admin the SPEC calls for:
tenant overview, licence search + inspect, journal head + chain
verify result, revocation list state. It is an MCP client of the core
(T0 read tools over the stdio face) with no direct store access and no
write path — v1 is read-only, firm-side only, bound to loopback
(systemd unit `cyan-eclipse-license-admin.service`, 127.0.0.1:8798 on
Shard E). Status: built 2026-11-12, benched **PASS** (2026-11-12),
tenant-#1 fixture scope.

TDQS

A4.1/5.0

Scored across 27 tools

Disambiguation4/5

The set is unusually well cross-referenced: every read tool explicitly names the sibling to use instead (license_status vs license_inspect vs license_verify vs license_list; seat_usage vs activation_list vs activation_inspect; chain_verify vs chain_verify_anchored), and the propose/execute split is clearly stated. The main residual ambiguity is license_renew vs license_supersede, which operate on the same renew-kind proposal but differ only by confirm verb and operation class — an agent could plausibly pick the wrong executor. revocation_list_get vs license_status is also a subtle boundary (published artifact vs live state), though the description warns about it.

Naming Consistency4/5

A consistent resource_action convention dominates (license_issue, license_revoke, journal_list, receipt_get, activation_inspect, product_register, tier_define), with a coherent _propose suffix marking the two-phase pattern against bare execute verbs. Deviations: revocation_list_get appends an extra noun-plus-'get' that breaks the pattern, and principal_manage is a multi-action tool hiding create/grant/revoke behind one name. Minor but noticeable irregularities.

Tool Count3/5

At 27 tools this is heavy, and the surface could be tightened (license_supersede overlaps license_renew; the journal has five read tools; propose/execute pairs double up four operation types). That said, the domain is genuinely broad — crypto-verified artefacts, two-phase issuance, append-only journal, activations, products/tiers, principals — so most tools do earn their place. It sits on the borderline of justified breadth rather than gratuitous sprawl.

Completeness4/5

Core lifecycle is fully covered: register product, define tier, propose/execute issue, renew, revoke, supersede, verify, list/inspect/status, revocation list, journal read and integrity checks, activation deactivation and stale-seat reclaim, plus principal/mandate admin. Notable gaps: there is no product or tier read tool (product_list/product_inspect, tier_list/tier_get) — after registering products and defining versioned tiers, an agent cannot enumerate or inspect them — and no activation-create path (plausibly client-driven). These are workable gaps but real ones.

Maintenance

ActivityMaintained
ResponsivenessNo recent community issues