Skip to main content
Glama

Eclipse Herald MCP

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

Eclipse Herald — the agent-to-agent discovery and delegation layer for Cyan Eclipse Technologies: signed Agent Cards, policy-gated delegation, certification before hire.

Why this exists

Eclipse Bazaar's Phase 3 is agent-to-agent commerce: agents finding, verifying, hiring, and paying each other. The wire protocol for that already exists as an open standard — A2A (Agent2Agent). Herald is CETech's implementation stance on it: the herald proclaims who our agents are (signed cards), carries our mandates outward (the gateway), and announces who else may be trusted (certify-before-you-hire). Blueprint: ~/workspace/cyan-eclipse/eclipse-herald-a2a-blueprint-20261025.md.

Related MCP server: @sequesign/mcp

Status

v0.1.0 — H0/H1 local implementation. In development; bench pending; NO certification claimed. Local only: no network calls, no live third-party agents, no real money (settlement is in zero-value TEST-CREDITS). Cards are signed and verifiable; delegation runs against local stub agents. Live task endpoints and external counterparties are H2/H3 scope.

What v0.1 does

  • Agent Card toolkit — builds A2A v1.0-shaped cards for the four hireable service faces (Test Bench, Verity audit, Argus research, Silas engineering) from card_templates.json; validates schema; signs with a locally generated RSA key (JWS/RS256, detached signature in the card's signatures array); exports the public JWKS. Signed cards for all four faces live in cards/; the public key document is keys/jwks.json.

  • Key custody — the private key is a single local file, keys/signing-key.json, chmod 600, generated on first use. It is never printed, logged, or embedded in any document. (Owner: keep a backup in your password manager, same pattern as the Memory Vault key.)

  • Gateway core — discover (local card file) → verify JWS → certify-check (card must carry a CERTIFIED verdict reference whose scope covers the requested skill) → mandate (AP2-shaped: task hash, counterparty card hash, budget cap, expiry) → dispatch to a local stub agent → delivery validation → receipt-on-delivery: the settlement record is written only when an artifact actually arrives. No artifact, no payment.

  • Security Law #1, enforced in the code path — foreign-agent output INFORMS; it never INSTRUCTS. Deliverables pass through a quarantine ingest that strips instruction-shaped lines into a flagged list and wraps the remainder as labeled data with provenance. A test proves an injected instruction in a deliverable changes nothing about gateway behavior.

  • Receipts ledger — append-only, hash-chained JSONL (receipts.jsonl): mandates, refusals, deliveries. The chain is verifiable; tampering breaks it detectably.

Tools (5, MCP + CLI parity)

Tool

What it does

card_build

Build an unsigned card for a face from its template.

card_sign

Build + sign a card; write it to cards/; export JWKS.

card_verify

Schema-validate a card and verify its JWS signature.

gateway_delegate

Governed delegation to a LOCAL test agent (echo, research-stub, silent, injecting).

receipt_list

Read the ledger + verify the hash chain.

Run it

python3 server.py                       # MCP over stdio, stdlib only
python3 herald_cli.py keygen            # one-time: signing key + JWKS
python3 herald_cli.py sign argus-research
python3 herald_cli.py verify cards/argus-research.agent-card.json
python3 herald_cli.py delegate cards/argus-research.agent-card.json research-stub research-brief 5.00 "Why did BTC move today?"
python3 herald_cli.py receipts
python3 -m unittest discover -s tests   # 17 tests

Env overrides: HERALD_KEY_PATH, HERALD_LEDGER, HERALD_CARDS_DIR, HERALD_JWKS.

Roadmap

  • H2 — Gateway dogfood against our own live service endpoints (Eclipse Connect relay pattern); internal worker-desk cards.

  • H3 — Bazaar P3: certify-before-you-hire as a paid service; external counterparties; real settlement behind the same mandate/receipt flow.

License / attribution

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

Available Tools

5 tools
card_buildB

Build an A2A v1.0-shaped Agent Card for a hireable CETech face (test-bench, verity-audit, argus-research, silas-engineering) from its template. Unsigned.

Inputs: face (required), overrides.

card_signB

Build + sign a face's Agent Card with the local signing key (JWS/RS256), write it to cards/, and export the public JWKS. The private key never leaves keys/ (chmod 600).

Inputs: face (required), overrides.

card_verifyA

Validate a card's schema and verify its JWS signature against the local JWKS. Tampered cards are rejected.

Inputs: card_path (required).

gateway_delegateB

Delegate a task through the governed gateway to a LOCAL test agent (echo, research-stub, silent, injecting): verify card -> certify-check -> mandate (budget cap, expiry) -> dispatch -> quarantine ingest -> receipt-on-delivery. No artifact, no payment.

Inputs: agent_id (required), skill_id (required), card_path (required), budget_cap_credits (required), +2 more.

receipt_listB

Read the hash-chained delegation ledger: mandates, refusals, delivery receipts. Includes chain verification.

Inputs: limit.

Tool Schema Changelog

Recent tool additions, removals, and schema changes observed during successful MCP inspections.

Loading tool schema changelog…

    TDQS

    A3.6/5.0

    Scored across 5 tools

    Disambiguation4/5

    Each tool targets a distinct stage of the card/delegation lifecycle, and gateway_delegate vs receipt_list are clearly separate concerns. The one soft overlap is card_build versus card_sign, since card_sign explicitly also builds the card, making card_build a strict subset step—descriptions disambiguate but an agent could reasonably pick the wrong one for a signed-card need.

    Naming Consistency5/5

    All names follow a uniform snake_case noun_verb pattern (card_build, card_sign, card_verify, gateway_delegate, receipt_list). The subject-first convention is applied consistently across the whole surface with no style mixing.

    Tool Count5/5

    Five tools is well-scoped for a narrow domain: card build/sign/verify, one delegation entry point, and one ledger read. Each tool maps to a real step in the pipeline with no filler or redundancy.

    Completeness4/5

    The core lifecycle (build → sign → verify → delegate → audit via receipts) is fully covered with no dead ends. Minor gaps remain: there is no tool to list/get existing cards from cards/, no key rotation or card revocation, and delegation is restricted to local test agents by design.

    Maintenance

    ActivityNo data
    ResponsivenessSyncing

    Related MCP Connectors

    Related MCP Servers

    • A
      license
      Not graded
      quality
      D
      maintenance
      Enables agent settlement, trust verification, and ledger operations for multi-agent workflows, with tools for blueprint management, credit tracking, and provenance recording.
      1
      MIT
    • A
      license
      A
      quality
      C
      maintenance
      Enables AI agents to create cryptographically verifiable receipts of their delegated work, with capabilities for multi-party approval and offline verification.
      11
      84 npm
      Apache 2.0
    • F
      license
      Not graded
      quality
      C
      maintenance
      Enables AI agents to discover other agents, publish and match tasks, exchange messages and artifacts, and build transaction-backed reputation over MCP and A2A.
      -