Skip to main content
Glama
mounesh00

Arqon SafeAction

by mounesh00

Arqon SafeAction

SOURCE RELEASE CANDIDATE — ready for publication authorization. MIT License: Copyright (c) 2026 Mounesh H Pattar. Owner attestations recorded; publication still requires separate authorization.

An assistant proposes. A person reviews. Only the exact approved checklist is saved.

This is a simulated Alexa+-style experience using an MCP-compatible architecture. It is not a live Alexa+ integration. All demonstration information is synthetic.

Problem and solution

Assistant-triggered actions need a clear boundary between proposing an action and performing it. SafeAction prepares a bounded travel checklist through the official MCP SDK. An independently authenticated local operator reviews its content and fingerprint, then explicitly approves saving it or rejects it without saving.

Related MCP server: com.aleeth/authority-mcp

Features

  • Exactly three MCP tools: arqon_health, arqon_run_mock, arqon_propose_checklist.

  • Deterministic mock output and five-item synthetic travel proposals.

  • Durable SQLite proposals with ten-minute expiry and atomic decisions.

  • Local operator login, HttpOnly SameSite=Strict session cookies and session CSRF.

  • Exact reviewed fingerprints; replay, tampering and unauthorized decision rejection.

  • Next.js review UI with separate confirmation and visible decision status.

  • No model credentials, real inference, scanner code, shell tools or Alexa+ connection.

Architecture and stack

Official MCP client → stdio server → SQLite ← FastAPI ← Next.js ← human review

Python 3.13.5, official MCP SDK 1.30.0, FastAPI 0.115.6, SQLite (stdlib), Node 22, Next.js 16.4.0, React 19.3.0 and TypeScript 5.9.3 (npm lockfile). See ARCHITECTURE.md and SECURITY.md.

Installation and configuration

The archive contains source, not environments or downloaded dependencies. Use the pinned separate Python profiles; the MCP and backend environments must not be merged. MCP's newer Pydantic/Starlette differ from this tested backend profile. Installation commands below are for the owner's later fresh setup; no dependencies were installed during release preparation.

python3.13 -m venv .venv-backend
python3.13 -m venv .venv-mcp
.venv-backend/bin/python -m pip install -r requirements-backend.txt
.venv-mcp/bin/python -m pip install -r requirements-mcp.txt
cd apps/web
npm ci
cd ../..

Run from this package root; no editable install is required. The stdlib application modules need no original Architect-VAPT directory. .env.example lists server-only configuration names and placeholders. Prefer the launcher: it generates backend credentials internally, prompts for a hidden independent password and creates private temporary storage. Never load placeholder values as real credentials.

Demo

.venv-mcp/bin/python -B -m examples.amazon_recording_demo

Wait for READY, then open the printed loopback URL. It creates fresh synthetic A/B proposals. Follow docs/AMAZON_DEMO.md and the 170-second video script. Ctrl-C or the instance's printed STOP command closes only owned services and removes its temporary data. No user approval is automated by the launcher.

For a production-build rehearsal, first run npm run build in apps/web, then add --production to the launcher. This remains local-only and is not a deployment.

Tests and production build

bash scripts/verify_local.sh

This runs only packaged relevant tests, TypeScript and the production build. It needs the preinstalled environments above. Tests use mocks or controlled local protocols; no live model invocation. See release evidence for exact measured counts and remaining owner gates.

Screenshots and recording

These screenshots come from the actual standalone production-build synthetic browser rehearsal. UUIDs and timestamps vary each run. No video is included or claimed.

Actual local production approval

Actual local production rejection

Record the actual three-minute demonstration before submission.

Limitations and release status

Local single operator and one persistent Node process; no public deployment, voice/Alexa+ registration, remote MCP endpoint, model provider or travel booking. Mock response history is not persisted. Demo data is deleted on launcher cleanup. Sessions are in-memory. Database write access and same-OS-user compromise are outside the application boundary. See release checklist.

MIT license and publication status

LICENSE retains the complete MIT text and Copyright (c) 2026 Mounesh H Pattar. The owner confirms original project code/publication permission, Arqon branding permission and screenshot/visual redistribution rights. These are owner attestations, not independently verified legal conclusions.

The scoped source-only review is READY FOR PUBLICATION AUTHORIZATION; see FINAL_RIGHTS_REVIEW.md. No identified shipped third-party notice/permission blocker remains. External dependencies are referenced, not bundled, and retain their own terms; future builds need fresh notice review.

Separate owner authorization is still required before GitHub publication. Intended repository: arqon-safeaction; prepared commands are in GITHUB_PUBLICATION_GUIDE.md. Nothing was published.

Related MCP Connectors

Related MCP Servers

  • F
    license
    Not graded
    quality
    D
    maintenance
    Turns OpenAPI specs into MCP tools with secure defaults, risk inspection, confirmation gates, response limits, audit logging, and secret redaction.
    -
  • F
    license
    Not graded
    quality
    D
    maintenance
    A governed MCP server exposing 37 risk-checked tools with signed Ed25519 receipts on a Bitcoin-anchored ledger, failing closed if governance or receipt writing fails.
    -
  • A
    license
    Not graded
    quality
    C
    maintenance
    Exposes standard MCP tools for secure calculation, knowledge-base retrieval, document statistics, and sensitive operations like email, export, and deletion with human-in-the-loop approval.
    MIT
  • F
    license
    Not graded
    quality
    B
    maintenance
    Enables MCP clients to discover and invoke tools for fictional customer lookup and simulated USD refunds, with strict argument validation, business-rule checks, and SQLite-backed audit persistence.
    -