Virtualize
Click on "Deploy Server".
Wait a few minutes for the server to deploy. Once ready, it will show a "Started" state.
In the chat, type
@followed by the MCP server name and your instructions, e.g., "@Virtualizecreate a VM with Ubuntu 22.04 and 4GB RAM"
That's it! The server will respond to your query, and you can continue using it as needed.
Here is a step-by-step guide with screenshots.
Virtualize
Free, cross-platform VM orchestration for AI workflows.
Virtualize gives AI agents (and humans) full VM lifecycle management with built-in MCP integration, sandboxed code execution, and a compliance-ready architecture (SOC 1/2/3, HIPAA, ISO 27001).
LLMs: Read
AGENTS.mdfor machine-readable project context, algebra definitions, tool chain format, and architecture invariants.

Why Virtualize?
Most AI workflows need sandboxed environments — to run generated code safely, test deployments, or give agents real OS-level access. Existing solutions are either cloud-locked, expensive, or platform-specific.
Virtualize is:
Free & open-source (Apache 2.0)
Cross-platform — Linux (KVM), macOS (Hypervisor.framework), Windows (WHPX/Hyper-V)
MCP-native — AI agents interact with VMs via the Model Context Protocol
Compliance-ready — audit logging, encryption, integrity chains, policy controls
Related MCP server: declaw-mcp-server
Architecture
┌──────────────────────────────────────────────────────────┐
│ AI Agents / Users │
├────────────┬────────────┬────────────┬───────────────────┤
│ MCP Server│ CLI │ REST API │ Web Dashboard │
├────────────┴────────────┴────────────┴───────────────────┤
│ VM Manager │
│ (orchestration + audit) │
├──────────────────────┬───────────────────────────────────┤
│ Sandbox Executor │ Compliance Engine │
│ (pooled isolation) │ (audit log + policy controls) │
├──────────────────────┴───────────────────────────────────┤
│ Hypervisor Abstraction │
│ ┌──────────┬──────────────┬──────────────┐ │
│ │ QEMU/KVM │ HVF (macOS) │ WHPX (Win) │ │
│ └──────────┴──────────────┴──────────────┘ │
└──────────────────────────────────────────────────────────┘Features
VM Management
Create, start, stop, destroy VMs with configurable CPU, memory, disk, network
GPU passthrough (VFIO on Linux) and virtual GPU support
Cloud-init support for automated provisioning
NAT, bridge, isolated, and host networking modes
Pre-built image support with copy-on-write overlays
MCP Server (for AI Agents)
13 tools exposed via the Model Context Protocol
vm_create,vm_start,vm_stop,vm_destroy— full lifecyclevm_exec— run commands inside VMssandbox_run— one-shot isolated code executionvm_file_read,vm_file_write— filesystem accesscompliance_report,audit_query,audit_verify— compliance tools
Sandboxed Code Execution
Run code in isolated VMs with strict resource limits
Timeout enforcement, CPU/memory caps
Pre-warmed VM pool for fast execution
Supports Python, Bash, Node.js, Ruby, Perl
Compliance
SOC 1/2/3 — Trust Services Criteria controls
HIPAA — 45 CFR § 164.312 audit and access controls
ISO 27001 — Annex A security controls
Immutable, integrity-chained audit logs (SHA-256 HMAC)
Optional encryption at rest (Fernet / AES-128-CBC)
Tamper detection with chain verification
Structured JSON logs for SIEM ingestion
Web Dashboard
Modern React UI with real-time VM monitoring
Create, start, stop, destroy VMs from the browser
In-browser terminal for VM command execution
Compliance report viewer
Formal Algebra
Virtualize is not just an MCP — it is an executable algebra. Every tool is a typed morphism over a formally defined state space, with verified axioms and constraint enforcement.
Classification
Virtualize MCP ≅ a typed, finite, partially-defined monoidal category with audit-preserving invariants
Structure
Component | Definition |
Carrier set |
|
Generators | 13 typed morphisms ( |
Composition |
|
Identity |
|
Constraint subalgebra |
|
Typed Transitions
Each tool is a morphism t_i : C_source → C_target with explicit preconditions:
vm_create : vm.nonexistent → vm.created
vm_start : vm.created | vm.stopped → vm.running
vm_stop : vm.running | vm.paused → vm.stopped
vm_destroy : vm.created | vm.running | vm.stopped | vm.paused → vm.destroyed
vm_exec : vm.running → vm.running (endomorphism)Verified Axioms
$ virtualize algebra verify
PASS identity — id ∘ t = t = t ∘ id holds for all generators
PASS closure — All generators map C → C
PASS associativity — (t₁ ∘ t₂) ∘ t₃ = t₁ ∘ (t₂ ∘ t₃)
PASS audit_monotonicity — A_{n+1}.seq ≥ A_n.seq
PASS audit_irreversibility — ∄ t such that t(A_n) = A_{n-1}
PASS transition_determinism — All transitions are deterministicKey Properties
Non-commutative:
create ∘ start ≠ start ∘ create(proven in tests)Audit chain:
A_{n+1} = H(A_n ∥ e_n)— monotonic, irreversible, append-onlyConstraint subalgebra: Compliance policies define
T_valid ⊆ T*(e.g., SOC2 blocks file reads when audit is tampered)Algebraic rewriting: Identity elimination, idempotent collapse, annihilation (
create ∘ destroy = id), dead code elimination
Plan Validation
Validate execution plans before running them:
# Valid lifecycle
$ virtualize algebra validate '[
["vm_create", null, {"name": "my-vm"}],
["vm_start", "my-vm", {}],
["vm_exec", "my-vm", {"command": "echo hello"}],
["vm_stop", "my-vm", {}],
["vm_destroy", "my-vm", {}]
]'
# → VALID — 5 steps validated
# Invalid: exec on nonexistent VM
$ virtualize algebra validate '[["vm_exec", "ghost", {}]]'
# → INVALID — vm_exec requires VM 'ghost' in {vm.running}, but it is in 'vm.nonexistent'Plan Optimization
$ virtualize algebra rewrite '[
["identity", null, {}],
["vm_create", "vm-1", {"name": "vm-1"}],
["identity", null, {}],
["vm_start", "vm-1", {}],
["vm_status", "vm-1", {}],
["vm_status", "vm-1", {}],
["vm_destroy", "vm-1", {}]
]'
# → Original: 7 steps → Optimized: 4 steps (3 eliminated via algebraic laws)Natural Language Agent
Ask in plain English — a small local LLM (Qwen 2.5 1.5B, ~1GB) translates your request into an algebraically validated tool chain. The algebra guarantees safety: invalid plans are rejected before touching any VM.
# Install agent dependencies
pip install -e ".[agent]"
# Ask anything
virtualize ask "start me a vm that i can connect to openclaw"Output:
╭─────────────── Execution Plan ───────────────╮
│ │
│ 1. Create VM 'openclaw-vm' │
│ 2. Start VM on 'openclaw-vm' │
│ 3. Run `pip install openclaw && python -m │
│ openclaw` on 'openclaw-vm' │
│ │
╰───────────────────────────────────────────────╯
VALID — 3 steps, audit seq → 3Add --execute (-x) to actually run the plan. Use --gpu-layers 0 for CPU-only inference.
How it works
User (English) → LLM → JSON tool chain → Compositor.validate() → Execute
↓ (if invalid)
Retry with error feedbackThe LLM can hallucinate any plan it wants — the algebra's compositor validates every step against the typed transition rules before execution. Invalid plans are fed back to the LLM with the specific algebraic violation for self-correction (up to 2 retries).
More examples
virtualize ask "create a vm called dev-box"
virtualize ask "check compliance for hipaa"
virtualize ask "make a vm, start it, and run uname"
virtualize ask "run print(42) in a sandbox"Quick Start
Prerequisites
The easiest way — let Virtualize detect your OS and install everything:
pip install -e .
virtualize setupThis will detect your OS, distro, package manager, hardware acceleration, and GPU — then install QEMU with the correct commands for your platform.
Or install manually:
# Linux (Ubuntu/Debian)
sudo apt install qemu-system-x86 qemu-utils
# Linux (Fedora/RHEL)
sudo dnf install qemu-system-x86 qemu-img
# macOS
brew install qemu
# Windows
choco install qemu
# or download from https://qemu.org/downloadInstall Virtualize
pip install -e .CLI Usage
# Create a VM
virtualize create my-dev-vm --cpus 4 --memory 4096 --disk 50
# Start it
virtualize start <vm_id>
# Run a command inside
virtualize exec <vm_id> "uname -a"
# Run sandboxed code
virtualize sandbox run "print('hello from sandbox')" --lang python
# List VMs
virtualize list
# Stop and destroy
virtualize stop <vm_id>
virtualize destroy <vm_id>API Server + Web Dashboard
# Start the API server (includes dashboard at http://localhost:8420)
python -m uvicorn virtualize.api.server:app --host 0.0.0.0 --port 8420MCP Server (for AI Agents)
Add to your MCP client configuration:
{
"mcpServers": {
"virtualize": {
"command": "python",
"args": ["-m", "virtualize.mcp_server.server"]
}
}
}Or start via CLI:
virtualize mcp serveCompliance
# Generate a SOC 2 compliance report
virtualize compliance report soc2
# Verify audit log integrity
virtualize compliance audit-verify
# Query audit events
virtualize compliance audit-query --actor alice --limit 20API Reference
REST Endpoints
Method | Path | Description |
|
| Web dashboard |
|
| Health check |
|
| Create VM |
|
| List VMs |
|
| Get VM details |
|
| Start VM |
|
| Stop VM |
|
| Destroy VM |
|
| Execute command in VM |
|
| Sandboxed code execution |
|
| Read file from VM |
|
| Write file to VM |
|
| Compliance report |
|
| List controls |
|
| Query audit log |
|
| Verify audit integrity |
|
| System information |
MCP Tools
Tool | Description |
| Create a new VM |
| Start a VM |
| Stop a VM |
| Destroy a VM |
| List all VMs |
| Get VM status |
| Execute command in VM |
| Isolated code execution |
| Read file from VM |
| Write file to VM |
| Generate compliance report |
| Query audit events |
| Verify audit log integrity |
Development
# Install with dev dependencies
pip install -e ".[dev]"
# Run tests
pytest
# Lint
ruff check src/ tests/
# Type check
mypy src/Project Structure
virtualize/
├── src/virtualize/
│ ├── core/
│ │ ├── algebra.py # Formal algebra: states, transitions, compositor, axioms
│ │ ├── models.py # Pydantic data models (VMConfig, VMInstance, AuditEvent)
│ │ ├── manager.py # VM lifecycle orchestration with algebraic pre-validation
│ │ ├── hypervisor.py # Cross-platform QEMU abstraction (KVM/HVF/WHPX)
│ │ ├── mock_hypervisor.py # Mock backend for dev/testing without QEMU
│ │ └── bootstrap.py # OS-detecting setup system
│ ├── agent/
│ │ └── nl_agent.py # NL→algebra agent (local LLM → validated tool chains)
│ ├── sandbox/
│ │ └── executor.py # Sandboxed code execution with pooled VMs
│ ├── compliance/
│ │ ├── audit.py # Append-only, integrity-chained audit log (SHA-256 HMAC)
│ │ └── policies.py # SOC 1/2/3, HIPAA, ISO 27001 policy controls
│ ├── mcp_server/
│ │ └── server.py # MCP server — 13 tools over stdio transport
│ ├── api/
│ │ ├── server.py # FastAPI REST server (port 8420)
│ │ └── dashboard.py # Built-in React/Tailwind web dashboard
│ └── cli/
│ └── main.py # Typer CLI (lifecycle, sandbox, compliance, algebra, ask)
├── tests/ # 103 tests (algebra, agent, API, compliance, models)
├── AGENTS.md # Machine-readable context for LLMs
├── bootstrap.sh # One-line clone + setup script
├── mcp-config.json # MCP client configuration
├── pyproject.toml
└── README.mdLicense
Apache License 2.0
This server cannot be deployed
Maintenance
Related MCP Connectors
Free OpenAI-compatible inference with signed provenance receipts and 3 focused MCP tools.
Hosted MCP catalog with 30 tenant-isolated browser, RAG, AI, mail and media tools.
MCP Hub: AI service discovery, per-user OAuth, and multi-service workflow orchestration
Compliance frameworks (SOC 2, ISO 27001, CMMC, NIST, more) delivered to AI agents as MCP tools.
Related MCP Servers
- AlicenseAqualityDmaintenanceAI workflow orchestration platform with 100+ integrations, knowledge graph, and unified credits. Build, manage, and execute workflows from Claude Code, Cursor, or any MCP client.40262 npm3MIT

declaw-mcp-serverofficial
AlicenseAqualityCmaintenanceRuns AI-generated code in secure Firecracker microVMs with opt-in network policy enforcement, PII scanning, prompt injection defense, and audit logging. Exposes MCP tools for running commands, managing files, and the full sandbox lifecycle.727 npm1Apache 2.0
vibops-mcpofficial
AlicenseAqualityAmaintenanceVibOps MCP is the control plane between your AI agents and your GPU infrastructure. 74 tools covering: GPU fleet management (deploy, scale, monitor across NVIDIA, AMD, Intel, AWS, Google, Groq), Agent Infrastructure Control Plane (per-agent GPU cost, budget enforcement, model policies, dependency graph), governance (AI Act, SOC 2, immutable HMAC audit chain), and GPU FinOps (chargeback, waste..)..7418MIT- AlicenseNot gradedqualityBmaintenanceMulti-model AI orchestration platform with 64 tools across 12 AI providers, enabling intelligent workflows from any MCP client.175 npm14AGPL 3.0