Delentia Sovereign AI Operating System

Registre officiel IA et agents

Advisory MCP tools: policy score, reasoning checklist, compression. To enforce, use delentia-guard.

Installation

Ajoutez ceci à la configuration MCP de votre client (Claude Desktop : claude_desktop_config.json ; Cursor : .cursor/mcp.json ; VS Code et les autres utilisent la même forme « mcpServers »).

NPM package delentia-mcp
Claude Code
claude mcp add delentia-sovereign-ai-operating-system -- npx -y delentia-mcp@2.1.2
Configuration JSON (Claude Desktop, Cursor, VS Code)
{
    "mcpServers": {
        "delentia-sovereign-ai-operating-system": {
            "command": "npx",
            "args": [
                "-y",
                "delentia-mcp@2.1.2"
            ]
        }
    }
}

Documentation

Tirée du README du projet sur GitHub, actualisée chaque semaine. Les liens ouvrent GitHub.

Sommaire
  1. Delentia MCP connector
  2. 🏛️ Executive Architectural Overview
  3. 🛠️ The 6 Tools
  4. 🚀 Quickstart: Universal 1-Click Client Setup
  5. 🌐 Official Marketplace & Registry Listings
  6. 🔒 Enterprise Security & Verification
  7. 🧩 Source code
  8. 📜 Intellectual Property & Attribution

Delentia MCP connector

Advisory policy-check, reasoning-checklist and context-compression tools for MCP agents, plus a pointer to Delentia Guard (the enforcing proxy)

What this package is: a small stdio bridge that lets MCP clients (Claude, Cursor, VS Code, Windsurf, Codex) call six tools hosted on Delentia's Cloudflare Workers. The tools are advisory: an agent that does not call evaluate_fdia is not stopped by it. To stop tool calls before they run, use Delentia Guard (npx delentia-guard), a local proxy that checks every call and keeps a signed, hash-chained log.
📊 Measured results, their limits, and the figures we withdrew


🏛️ Executive Architectural Overview

The hosted tools give an agent a deterministic policy score (F = D^I x A, same inputs, same answer), a seven-step reasoning checklist, deterministic context compression and signed task packets. They are checks an agent can ask for, not guarantees:

       ┌─────────────────────────────────────────────────────────────┐
       │              Autonomous AI Clients & IDEs                   │
       │    (Claude Desktop, Cursor, VS Code, Windsurf, Cline)       │
       └──────────────────────────────┬──────────────────────────────┘
                                      │
                         stdio / Streamable HTTP (MCP)
                                      │
       ┌──────────────────────────────▼──────────────────────────────┐
       │              Delentia Sovereign Edge Gateway                │
       │         (Cloudflare Global Workers + Durable Objects)       │
       └──────────────────────────────┬──────────────────────────────┘
                                      │
         ┌────────────────────────────┼────────────────────────────┐
         ▼                            ▼                            ▼
  [ 1. FDIA Security ]        [ 2. RCT-7 Thinking ]        [ 3. Delta & JITNA ]
Deterministic policy gate     7-stage reasoning            Context compression
    F = (D^I) * A             checklist                    & signed task packets

🛠️ The 6 Tools

Tool NameTypeWhat it does
evaluate_fdiaRead-OnlyScores a proposed action with F = (D^I) × A against your policy and returns AUTHORIZED or a block verdict with a SHA-256 audit digest. A = 0 (policy says no, or a required human signature is missing) always gives F = 0. Actions that need a human require an Ed25519-signed Architect token from a key the deployment trusts.
configure_policyActionSets the rules for your session only (RBAC, blocked actions, dual sign-off, safety threshold). The deployment's default policy is never changed.
rct_thinkRead-OnlyWalks a 7-stage reasoning checklist over a plan and returns a heuristic alignment score. It helps catch vague or misaligned plans; it does not guarantee correct answers.
compress_contextRead-OnlyDeterministic compression of long tool output and logs (no LLM call). Measured ~70% fewer tokens on real logs while literal-wording questions stayed answerable (benchmark). Can keep the original for expand_context.
expand_contextRead-OnlyReturns lines of the original behind a compressed result, by line range or search, when the compressed version left something out.
orchestrate_swarmRead-OnlySplits a goal into signed JITNA task packets for several roles.

Advisory vs enforcing: these tools help an agent that chooses to call them. To enforce a policy on every tool call of another MCP server, use Delentia Guard (delentia-guard), an stdio proxy with a signed audit log.


🚀 Quickstart: Universal 1-Click Client Setup

Universal Remote Bridge (Zero-Friction 1-Click Setup)

Connect any MCP-compatible environment (Claude Desktop, Cursor, Antigravity IDE, VS Code, Windsurf) directly to the Delentia Sovereign Global Edge:

npx -y mcp-remote https://delentia-sovereign-mcp.delentia.workers.dev/mcp

Developer Sandbox Free Tier: Includes 50 daily free calls per caller IP out of the box with zero registration required!
The count is kept in memory by each Worker instance, so it is a best-effort limit (it can reset when an instance restarts). Past the limit a tools/call returns HTTP 429 with JSON-RPC error code -32002; tools/list and initialize are not counted.
Paid tier (preview): For a higher quota, pass a Zuplo API Key from the Delentia Developer Portal:

npx -y mcp-remote https://delentia-sovereign-mcp.delentia.workers.dev/mcp --header "Authorization: Bearer YOUR_ZUPLO_API_KEY"
Option A: Cursor IDE Configuration (~/.cursor/mcp.json)
{
  "mcpServers": {
    "delentia-sovereign": {
      "command": "npx",
      "args": [
        "-y",
        "mcp-remote",
        "https://delentia-sovereign-mcp.delentia.workers.dev/mcp"
      ]
    }
  }
}
Option B: Claude Desktop Configuration (claude_desktop_config.json)
1. Free Community Sandbox (Zero-Config / 50 calls daily):
{
  "mcpServers": {
    "delentia-sovereign": {
      "command": "npx",
      "args": [
        "-y",
        "mcp-remote",
        "https://delentia-sovereign-mcp.delentia.workers.dev/mcp"
      ]
    }
  }
}
2. Paid tier (Zuplo API Key, higher quota — no SLA is offered yet):
{
  "mcpServers": {
    "delentia-sovereign": {
      "command": "node",
      "args": [
        "./bin/cli.js"
      ],
      "env": {
        "DELENTIA_API_KEY": "YOUR_ZUPLO_API_KEY"
      }
    }
  }
}
Option D: Instant Terminal Verification (1-Line cURL)

Test the live Sovereign MCP server directly in any terminal:

curl -X POST https://delentia-sovereign-mcp.delentia.workers.dev/mcp \
  -H "Content-Type: application/json" \
  -d "{\"jsonrpc\":\"2.0\",\"id\":1,\"method\":\"tools/list\"}"

🌐 Official Marketplace & Registry Listings

Marketplace / DirectoryStatusOfficial Live Listing Link
Smithery.aiPublishedsmithery.ai/servers/delentia/delentia-sovereign
Glama.aiPublishedglama.ai/mcp/servers/delentia-labs/delentia-mcp
Official MCP RegistryPublishedregistry.modelcontextprotocol.io/?q=delentia
MCPizePublishedmcpize.com/mcp/delentia-mcp
Zuplo Developer PortalPublisheddelentia-gateway-main-c7624a5.zuplo.site/introduction
PulseMCPAuto-SyncingSynced via Official MCP Registry Index
Awesome MCP ServersPR Submitted (Pending Merge)github.com/punkpeye/awesome-mcp-servers/pull/13430
MCP Market (CherryHQ)Issue Submitted (Pending Review)github.com/CherryHQ/mcpmarket/issues/51

🔒 Enterprise Security & Verification

  • Audit digest: every evaluation returns a SHA-256 digest of its inputs and verdict. A digest shows that a record was changed; on its own it does not make a log tamper-proof. For an enforcing, Ed25519-signed, hash-chained log of every tool call, run Delentia Guard.
  • Architect sign-off: high-risk actions need an Ed25519-signed Architect token bound to that action and payload, from a key the deployment trusts; dual sign-off actions need two different trusted keys. Unsigned or forged tokens fail closed.
  • Structured pre-execution reasoning: rct_think walks a 7-stage checklist and returns a heuristic alignment score. It helps catch vague or misaligned plans; it does not guarantee the absence of hallucinations.

🧩 Source code

This repository holds both the public connector (root: bin/, dist/, listing files) and the server source in ecosystem/: the Cloudflare Workers behind the 6 tools (sovereign, fdia, rct7, delta, jitna, intent-loop, shared) and Delentia Guard (packages/guard).

cd ecosystem
npm ci
npm run build && npm run typecheck && npm run test:all

Cross-stack bridge tests that need Delentia's non-public Python services skip unless DELENTIA_PRIVATE_SERVICES_DIR is set.


📜 Intellectual Property & Attribution

  • Developer: Delentia Labs
  • Chief Architect: Ittirit Saengow (The Architect)
  • Official Portal: https://delentia.com
  • License: Apache-2.0 (connector and ecosystem/ server source)

Questions fréquentes

Qu'est-ce que Delentia Sovereign AI Operating System ?

Advisory MCP tools: policy score, reasoning checklist, compression. To enforce, use delentia-guard.

Comment installer Delentia Sovereign AI Operating System ?

Il s'exécute sur votre machine. Dans Claude Code : claude mcp add delentia-sovereign-ai-operating-system -- npx -y delentia-mcp@2.1.2. Pour les autres clients, collez la configuration JSON de cette page dans leurs réglages MCP.

Faut-il une clé API pour Delentia Sovereign AI Operating System ?

Son entrée au registre n'indique aucune clé ni secret obligatoire. Certaines fonctions peuvent tout de même demander un compte auprès du service auquel il se connecte.

Delentia Sovereign AI Operating System est-il open source ?

Oui. Son code est sur GitHub sous licence Apache-2.0.

Qui maintient Delentia Sovereign AI Operating System ?

Il est publié par delentia-labs. AI.tn le référence sans le maintenir ni le recommander.

Issu du registre MCP officiel. Être listé n'est pas une recommandation : vérifiez le code et les permissions d'un serveur avant de lui donner accès à vos données.