AGTS MCP Server

io.github.ObligationSign由 GitHub 帳號 ObligationSign 發布v1.0.1更新於 Oct 24, 2026

Sovereign MCP Gateway — 64 governed tools with Merkle transparency log. EU AI Act and SOC 2 ready.

發布方自己的服務:AGTS MCP Server

已驗證Streamable HTTP可網頁執行雲端與基礎設施知識與記憶安全與監控

概覽

AI 產生的概覽

遠端治理閘道,提供 115 個以上工具,讓代理動作先經過五重門防火牆評估,並記錄到簽章且以 Merkle 錨定的透明日誌中。

可以做什麼
治理:提交動作進行執行前閘控評估、以 Merkle 證明驗證承諾、回報執行結果、管理政策與統計。身分與信任:註冊代理護照、掃描技能、以抗量子密碼學認證對等方並查詢信任狀態。郵件與雲端硬碟:傳送加密郵件、讀取收件匣,以及上傳、列出、共享、下載或刪除檔案。記憶:儲存、召回、搜尋和整併階層式記憶項目,支援 TTL 與垃圾回收。計費與編排:管理支出與預算,派生、串接、輪詢、取消和合併子代理及工作流程範本。基礎設施:管理 VPN 通道、DNS 查詢、健康探測與告警。
適用情境
當你需要可稽核地記錄 AI 代理被授權做什麼、實際做了什麼時適用,例如受監管或合規敏感的工作流程;也適合希望在一個閘道後取得受治理的郵件、加密檔案儲存、代理身分和子代理編排的情境。
執行需求
遠端 Streamable HTTP 端點,無需本機執行環境。需要在 Authorization 標頭中提供由 ObligationSign 平台簽發的 Bearer 權杖,須先在供應商網站註冊取得。需要能存取該閘道的網路。
安裝前請注意
閘道需要帳號和 Authorization Bearer 權杖,屬於需妥善保管的憑證。工具可傳送郵件、上傳、共享和刪除檔案、刪除記憶、變更政策、凍結支出並啟動子代理,寫入和刪除會實際生效。計費工具涉及預算與支出管理。內容與動作會經過第三方服務及其透明日誌。

安裝

在 SourceWeft 中

  1. 開啟 儀表板中的 AGTS MCP Server,將其新增到工作區。
  2. 為需要使用其工具的對話啟用該服務。

Web executable,透過 Streamable HTTP。 遠端服務在工作區中設定後即可從網頁執行環境執行。

交給代理程式安裝

把這段提示詞貼上到 Claude Code、Codex、Cursor 等代理程式,或 SourceWeft 對話中。代理程式會讀取該伺服器的安裝指南,先向你展示發布者、端點和登入方式,經你同意後再把它新增到你的客戶端。金鑰不會寫進任何命令。

閱讀 https://sourceweft.com/mcp/io.github.ObligationSign%2Fagts-mcp/install.md,並按其中說明把這個 MCP 伺服器接入我的客戶端。修改任何設定之前,先告訴我它的發布者、端點和登入方式,等我確認。不要把 API 金鑰寫進命令或我會提交的檔案;問我金鑰應從哪裡取得。

檢視代理程式所依據的安裝指南

其他 MCP 客戶端

把它新增到你客戶端的 mcpServers 設定中。

{
  "mcpServers": {
    "agts": {
      "type": "http",
      "url": "https://agts-mcp.obligationsign.com/mcp"
    }
  }
}

README

AGTS MCP Server

Sovereign MCP Gateway — Governed Tool Invocation for Autonomous Agents

Model Context Protocol server that turns tool calls into cryptographically authorized, verifiable actions. 115+ governed tools across 6 layers — hybrid Ed25519 + SLH-DSA signed and Merkle-anchored. Designed to support EU AI Act, GDPR, SOC 2, ISO 27001, DORA, NIS2, and FINRA compliance.


Server URL

https://mcp.obligationsign.com/mcp
TransportURLStatus
Streamable HTTP (primary)https://mcp.obligationsign.com/mcpActive
SSE (deprecated)https://mcp.obligationsign.com/mcp/sseDeprecated

Authentication

The gateway requires a Bearer token issued through the ObligationSign platform. All requests must include an Authorization header.

To obtain a token, register at obligationsign.com/start/.

Quick Start

Claude Desktop / Cursor / Windsurf

Add to your MCP configuration:

json
{  "mcpServers": {    "agts": {      "url": "https://mcp.obligationsign.com/mcp",      "headers": {        "Authorization": "Bearer YOUR_TOKEN"      }    }  }}

Programmatic (JSON-RPC over HTTP)

bash
curl -X POST https://mcp.obligationsign.com/mcp \  -H "Content-Type: application/json" \  -H "Authorization: Bearer YOUR_TOKEN" \  -d '{    "jsonrpc": "2.0",    "id": 1,    "method": "tools/list",    "params": {}  }'

What is AGTS?

AGTS (Autonomous Governance Transparency Standard) is a transparency-backed governance protocol that converts validated machine decisions into append-only cryptographic records, independently verified by a network of witnesses and monitors.

decision → record → immutable history → verifiable truth

Every tool call passes through a five-gate firewall (Statistical, Causal, Regression, Evidence, Human-Compatible Explanation) before execution. Admitted actions are Merkle-anchored into an append-only transparency log and signed with hybrid post-quantum cryptography.


Cryptography & Validator Quorum

Every governance envelope produced by this server is signed with a hybrid post-quantum signature: classical Ed25519 in parallel with SLH-DSA-SHAKE-128f. The signature is verifiable today with Ed25519-only tooling and remains forgery-resistant against future quantum adversaries via SLH-DSA. Sovereign Mail additionally uses ML-KEM-512 for encrypted key exchange, and peer-trust ceremonies use ML-DSA-44.

Admit decisions for governed tool calls require an independent 3-of-4 validator quorum with constraint regulator_count ≥ 1 and auditor_count ≥ 1. Validators are deployed as separate Cloudflare Workers (agts-validator-{a,b,c,d}) with distinct ECDSA P-256 keypairs; each validator independently re-evaluates the proof bundle and signs an AGTS_VOTE_V1 certificate. The quorum certificate is then anchored alongside the governance envelope in the Merkle transparency log, so any external party can reconstruct the four signed votes and verify that quorum was honestly met.

Tool Catalog (115+ Tools)

Layer 1 — Infrastructure

VPN & Network
ToolDescription
create_tunnelEstablish an attested VPN tunnel (WireGuard/IPSec/TLS)
tunnel_statusMonitor uptime, transfer bytes, and attestation status
disconnect_tunnelTerminate session and log the event
dns_querySecure DNS resolution through the Sovereign resolver
Monitor
ToolDescription
system_healthProbe all mesh workers for status and latency
check_alertsFetch the monitor alert timeline
subscribe_alertsWebhook subscription for real-time alerts
check_equivocationsFetch conflicting Signed Tree Head proofs
gossip_statusCurrent state of gossip protocol and monitor identity

Layer 2 — Identity & Communications

Sovereign Mail
ToolDescription
send_sovereign_mailSend encrypted (internal) or plaintext (external) messages
list_inboxList message metadata from an inbox
read_messageRetrieve full message content by ID
lookup_recipient_keysRetrieve public keys (ECDH P-384 / ML-KEM-512)
Sovereign Drive
ToolDescription
upload_fileClient-side encrypted upload with integrity anchoring
list_filesList file metadata, sizes, and sharing status
download_fileRetrieve ciphertext for client-side decryption
share_fileShare files using ECDH key wrapping
delete_fileRemove a file and create a death leaf in the log
OpenClaw (Agent Identity)
ToolDescription
register_clawCreate a cryptographic Claw Passport and identity keys
scan_skillScan agent skills through the five-gate firewall
claw_statusReturn trust score, governance history, and quota
export_passportSecure export of the identity envelope
govern_skill_execHigh-level gate for skill execution
report_skill_executionLink skill results to authorization
verify_skill_proofCryptographic verification of skill scans
Peer Trust (Post-Quantum)
ToolDescription
authenticate_peerHybrid PQC (ML-DSA-44) + Classical (ECDSA) auth ceremony
verify_agent_identityFingerprint and revocation check via key registry
exchange_credentialsSecure transport using hybrid ECDH + ML-KEM-512
peer_trust_networkList known peers and mutual trust status

Layer 3 — Governance

ToolDescription
submit_governancePre-execution gate evaluation (ADMIT/QUARANTINE/REFUSE)
verify_commitmentVerify commitment in transparency log with Merkle proofs
report_executionLog outcome of an authorized action
update_policyUpdate firewall parameters
list_policiesList active governance thresholds and history
governance_reportAggregated decision reporting
governance_statsPer-gate pass/fail statistics and HCE distribution
gate_analysisDetailed breakdown of a specific evaluation decision

Layer 4 — Memory

ToolDescription
memory_storeStore hierarchical key-value entries
memory_recallRetrieve specific entries by path
memory_listPrefix-based listing of memories
memory_searchSemantic similarity search using vector embeddings
cross_session_recallSemantic search across historical sessions
memory_contextMode-based contextual summary
consolidate_memoryCompare and compact redundant entries
memory_gcGarbage collection of stale entries
set_memory_ttlConfigure automatic expiration for entries
memory_deleteRemove specific memory paths

Layer 5 — Billing

ToolDescription
provision_resourceCreate a billing project for an agent
manage_spendingFreeze/unfreeze spending
check_project_billingDetailed usage and charge breakdown
verify_provisionCheck billing project linkage
set_budget_alertConfigure spending thresholds
check_budgetReal-time spend vs. cap tracking

Layer 6 — Orchestration

ToolDescription
spawn_sub_agentLaunch an isolated asynchronous sub-agent
sub_agent_statusPoll sub-agent execution status
sub_agent_resultRetrieve sub-agent results
chain_delegationRecursive delegation with budget decay and depth limits
create_templateDefine a reusable multi-step workflow
apply_templateInstantiate a template into a running task
list_sub_agentsList active sub-agents
list_templatesList available workflow templates
cancel_sub_agentTerminate a running sub-agent
merge_sub_resultsAggregate results from parallel sub-agents
delegation_treeReconstruct the full lineage of a delegated chain

REST API

In addition to MCP, the gateway exposes REST endpoints:

EndpointDescription
GET /api/v1/healthUpstream service health checks
GET /api/v1/auditAudit log queries
GET /api/v1/assetsAsset management
GET /verify?commitment={hash}Commitment verification with Merkle proof

Architecture

The MCP server is one access surface to the AGTS protocol. The full clearinghouse specification is published separately.

Agent ─── MCP Gateway ─── Five-Gate Firewall ─── Transparency Log               │                                        │               ├── Mail Worker (E2EE)                   ├── Merkle Tree               ├── Drive Worker (Encrypted Storage)     ├── Signed Tree Heads               ├── OpenClaw (Identity)                  └── Gossip Protocol               ├── Tunnel Worker (VPN)               ├── Monitor Worker (Health)               └── Protocol Worker (Governance)

How This Server Differs

Most MCP servers in the ecosystem (~83%) are thin API wrappers — a single connector exposed as one or two tools, running locally via stdio, with no authentication or validation. The production-tier servers from vendors like GitHub, Slack, and AWS (~2%) are well-built but architecturally simple: they translate tools/call into API requests for their own service.

This server operates at a different level. It is not a wrapper around an external API — it implements a full governance protocol where:

  • Every action must pass through five evaluation gates before execution is authorized
  • Admitted actions are cryptographically signed (hybrid Ed25519 + SLH-DSA) and Merkle-anchored in an append-only transparency log
  • Execution outcomes are traced back to their authorization, creating a closed governance loop
  • Independent monitors verify log consistency through a gossip protocol, detecting equivocation or tampering
  • Identity is handled through post-quantum cryptography (ML-DSA-44, ML-KEM-512) for forward security

The result is that every tool call produces independently verifiable cryptographic evidence of what was authorized, what was executed, and whether the outcome matched the authorization — the kind of audit trail required in regulated industries (financial services, healthcare, legal, critical infrastructure).

Regulatory Readiness

The gateway's architecture is designed to support compliance across multiple regulatory frameworks. The table below maps specific requirements to the AGTS features that address them.

EU AI Act
RequirementAGTS Implementation
Art. 9 — Risk managementFive-gate firewall evaluates every action before execution; QUARANTINE decisions enable human intervention
Art. 12 — Record-keepingAppend-only Merkle transparency log with signed governance envelopes; automatic recording of every authorization and execution event
Art. 13 — TransparencyEvery commitment is independently verifiable with cryptographic inclusion proofs; governance decisions are explainable via the HCE (Human-Compatible Explanation) gate
Art. 14 — Human oversightGate thresholds are configurable; QUARANTINE/REFUSE decisions halt execution; policy updates are themselves governed and logged
SOC 2
Trust PrincipleAGTS Implementation
SecurityHybrid Ed25519 + SLH-DSA signing; post-quantum key exchange (ML-KEM-512); Cloudflare Access authentication with email binding; per-tool rate limiting
AvailabilityMulti-worker mesh architecture with independent health monitoring; gossip protocol detects node failures; system_health probes all upstream services
Processing IntegrityFive-gate firewall validates every action before execution; execution outcomes are traced back to authorization; variance detection on deviated outcomes
ConfidentialityEnd-to-end encrypted mail (ECDH P-384 key wrapping); client-side encrypted drive storage; attested VPN tunnels for network isolation
PrivacyNo plaintext storage of message content; cryptographic identity via Claw Passport; key registry with revocation support
ISO 27001
Control AreaAGTS Implementation
A.8 — Asset managementEvery file, message, and identity key is tracked with cryptographic hashes; death leaves record asset deletion
A.9 — Access controlBearer token authentication; Cloudflare Access email binding; per-agent identity via Claw Passport with trust scoring
A.10 — CryptographyHybrid classical + post-quantum signing (Ed25519 + SLH-DSA); PQC key exchange (ML-KEM-512); ECDH P-384 for mail encryption
A.12 — Operations securityImmutable transparency log; independent monitor workers verify log consistency via gossip protocol; equivocation detection
A.16 — Incident managementAlert subscription system; equivocation checks detect conflicting Signed Tree Heads; governance variance records track deviations
A.18 — ComplianceAll governance decisions, policy changes, and execution outcomes are cryptographically anchored; audit trail is independently verifiable
DORA (Digital Operational Resilience Act)
RequirementAGTS Implementation
Art. 6 — ICT risk managementFive-gate firewall performs pre-execution risk evaluation; gate thresholds are configurable per policy; governance stats provide per-gate pass/fail analytics
Art. 9 — DetectionMonitor workers continuously probe service health; gossip protocol detects inconsistencies; alert subscriptions enable real-time incident notification
Art. 10 — Response and recoveryQUARANTINE decisions halt risky operations; manage_spending can freeze agent budgets; execution tracing links failures to their authorization
Art. 11 — TestingEvery governance decision is independently verifiable via Merkle inclusion proofs; verify_commitment enables third-party audit at any time
Art. 15 — Third-party riskSub-agent delegation enforces budget decay and depth limits; chain_delegation tracks recursive delegation lineage; delegation_tree reconstructs full provenance
GDPR
RequirementAGTS Implementation
Art. 5(1)(a) — Lawfulness, fairness, transparencyEvery agent action is pre-authorized through the five-gate firewall; governance decisions include HCE (Human-Compatible Explanation) for explainability
Art. 5(1)(e) — Storage limitationmemory_gc garbage collects stale entries; set_memory_ttl enforces automatic expiration; delete_file creates cryptographic death leaves proving deletion
Art. 5(1)(f) — Integrity and confidentialityEnd-to-end encrypted mail and drive storage; hybrid post-quantum cryptography; append-only transparency log ensures integrity
Art. 17 — Right to erasurememory_delete removes specific memory paths; delete_file with death leaf provides verifiable proof of deletion
Art. 25 — Data protection by designCryptographic identity via Claw Passport (no PII required); client-side encryption for stored files; key registry with revocation support
Art. 30 — Records of processing activitiesMerkle transparency log records every authorization, execution, and policy change with timestamps and cryptographic signatures
NIS2
RequirementAGTS Implementation
Art. 21 — Cybersecurity risk managementGovernance firewall, cryptographic signing, transparency log, and post-quantum key exchange form a defense-in-depth architecture
Art. 23 — Incident reportingAlert system with webhook subscriptions; equivocation detection; governance variance records provide forensic evidence chain
Art. 21(2)(d) — Supply chain securitySub-agent orchestration with budget controls and depth limits; delegation tree provides full provenance tracking across agent boundaries
FINRA
RequirementAGTS Implementation
Rule 3110 — SupervisionFive-gate firewall enforces pre-execution supervision of every agent action; QUARANTINE/REFUSE decisions prevent unauthorized transactions
Rule 3120 — Supervisory control systemGovernance stats and gate analysis provide supervisory analytics; policy thresholds are configurable and themselves governed
Rule 4511 — Books and recordsAppend-only Merkle transparency log with Ed25519 + SLH-DSA signed governance envelopes; every authorization and execution is permanently recorded
Rule 3310 — AML compliancesubmit_governance evaluates evidence including value, recipient, and intent before any financial action is authorized; variance detection flags deviations
Rule 2210 — Communicationssend_sovereign_mail with governance pre-authorization ensures all agent communications are reviewed and logged before transmission

The closest comparable in the market is not another MCP server — it is what a regulated enterprise would build internally to control what their AI agents can do in production.


Specification

The normative AGTS Clearinghouse specification is available at:

github.com/obligationsign/agts-clearinghouse


Links


License

This document is published for reference and integration purposes. The MCP server is operated by ObligationSign. See the AGTS Clearinghouse Specification for protocol terms.

來源:README.md,提交 b690655

工具

0
SourceWeft 尚未讀取此伺服器的工具。

版本歷史

1
  1. v1.0.1最新Sep 16, 2026

評分

已滿足適用檢查項的 40%

該評分衡量此收錄條目的完整程度,依據註冊中心條目、端點與公開來源計算得出。它不是對伺服器本身的評價,也不表示伺服器安全或目前可用:可連通性是帶日期的檢查結果,評分也不包含安裝量或使用者評價。

  • 近 30 天內可連通必需25 分

    請保持端點能夠回應 MCP 握手;SourceWeft 每 30 天探測一次。

  • 已宣告認證方式必需10 分

    請在註冊中心條目中宣告憑證請求標頭,或保持端點能夠回應探測,以便判定其認證方式。

  • 已列出工具15 分

    請讓伺服器在無憑證時回應 tools/list,或發布可在沙箱中無憑證啟動的軟體包。

  • 描述必需15 分
  • 圖示10 分
  • 文件15 分
  • 授權條款10 分

    請在儲存庫或註冊中心條目中宣告授權條款。

其他 Obligationsign 整合