Skip to main content
Glama

Varco — the approval gateway for AI agents

Your AI agents act on real business systems. You stay in control.

Varco sits between any MCP-compatible agent (Claude, Copilot, Cursor, custom agents) and your backend (ERP, or any REST/Odoo system) and enforces:

  • Per-agent permissions — each agent has an identity and sees only what it's allowed to; unregistered agents are rejected at the first tool call.

  • Autonomy thresholds — writes within policy (e.g. orders under €1,000) execute immediately and are logged; everything else waits for a human. Conservative by design: no threshold, no readable amount → always ask.

  • Human approval with review-and-edit — approvers see the request in plain language, can fix values before approving, and decide with one tap from the web app or Telegram (WhatsApp next). Bulk approve, scheduled digests (TELEGRAM_DIGEST), and a second web-only threshold (soglia_web) for high amounts: no one-tap above it.

  • Multiple approvers with per-department delegationVARCO_APPROVERS (name:key:departments;...): everyone sees everything, each approves only their delegated departments, every decision carries the approver's name.

  • Full audit trail with CSV export — every read, request, approval, rejection and auto-execution is recorded and exportable (/export/audit.csv). Built for EU AI Act art. 14-style human oversight.

AI agent ──MCP──▶ varco_mcp ──▶ SQLite state ◀── dashboard / Telegram ◀── human
                     │                                   │
                     └─ reads (audited)                  └─ writes ONLY after
                                 ▼                          policy or approval
                     backend: mock ERP · REST · Odoo (JSON-RPC)

Quickstart

python -m venv .venv
.venv/Scripts/python -m pip install -r requirements.txt   # mcp, httpx
.venv/Scripts/python test_varco.py                        # all suites: test_*.py
.venv/Scripts/python varco_dashboard.py                   # http://127.0.0.1:8420

The default config ships a mock ERP with sample data, four departments (Sales, Administration, Purchasing, Warehouse) and five agents — a full working demo with no external system.

Connect an agent (Claude Code example):

claude mcp add varco --scope project -e VARCO_AGENT=assistente-vendite -- <path>/.venv/Scripts/python <path>/varco_mcp.py

Related MCP server: AgentsGate

Configuration

Everything lives in varco_config.json: backend (mock, REST base URL, or "tipo": "odoo" — see varco_config.odoo.json for a ready-made Odoo mapping), entities, departments, agents with read/write permissions, soglie (autonomy thresholds) and soglia_web (web-only confirmation above this amount). Secrets never go in the config: API keys via ERP_API_TOKEN, Telegram via TELEGRAM_BOT_TOKEN + TELEGRAM_CHAT_ID, dashboard login via VARCO_ACCESS_KEY (open demo mode when unset).

Tests

Five self-contained suites, no external services required: test_varco.py (policy + approvals), test_dashboard.py (web app), test_varco_mcp.py (real MCP protocol round-trip), test_varco_rest.py and test_varco_odoo.py (backend adapters against local fake servers).

Docs in Italian

Product docs and the SMB demo script: README.it.md.

Roadmap

WhatsApp Business approvals · Odoo App Store module · hosted cloud (free / $99 / $399) · approval digests & bulk approve · per-approver web confirmation above a second threshold.

MIT licensed. Built in Italy 🇮🇹

Related MCP Connectors

Related MCP Servers

  • A
    license
    Not graded
    quality
    D
    maintenance
    Human-in-the-Loop authorization gateway for AI Agents. Securely pause MCP workflows and route high-risk actions to human approvers via Slack or Email.
    64 npm
    1
    MIT
  • A
    license
    Not graded
    quality
    A
    maintenance
    Enables AI agents to securely call MCP tools with risk scoring, checkpoints, rollback, and approval workflows.
    1 npm
    MIT
  • F
    license
    Not graded
    quality
    B
    maintenance
    Provides a secure MCP boundary for AI agents, intercepting and validating tool calls, redacting secrets, and requiring human approval for sensitive actions with a tamper-evident audit trail.
    -
  • F
    license
    Not graded
    quality
    C
    maintenance
    Enables controlled AI-agent access to enterprise-shaped tools with a deny-by-default gated write path, human approval, dry-run execution, and append-only audit logging.
    1
    -