nano-vm-mcp 0.4.2

MCP server for llm-nano-vm — run deterministic LLM programs via Model Context Protocol

Stateful MCP gateway for deterministic LLM workflows.
Governance-first. Replayable. Audit-complete.
Built on llm-nano-vm — the deterministic FSM execution kernel.

---

What nano-vm-mcp Is

nano-vm-mcp is an MCP gateway that turns the Model Context Protocol into a governance-bound execution environment. It wraps the llm-nano-vm execution kernel and exposes it to any MCP client — Claude Desktop, Claude Code, custom agents, or API callers — through stdio or SSE transport.

Most MCP servers expose stateless tools. nano-vm-mcp exposes stateful, governed, auditable workflows.

Capability	Typical MCP Server	nano-vm-mcp
Tool execution	✅	✅
Stateful workflows	❌	✅
Deterministic FSM	❌	✅
Replayable traces	❌	✅
Suspend / resume	❌	✅
Capability enforcement	❌	✅
Append-only audit trail	❌	✅
GDPR tombstoning	❌	✅
Inter-session idempotency	❌	✅

Core invariant: the gateway does not own execution logic — the FSM kernel does.

δ(S, E) → S'

  S  — current execution state
  E  — validated event
  S' — next deterministic state

---

Architecture

MCP Client
  → nano-vm-mcp (Gateway)
      → GovernedRunProgramHandler   ← PolicySnapshot, idempotency_key, CapabilityRef resolution
          → llm-nano-vm (Kernel)    ← deterministic FSM, ASTEngine, ProjectionLayer
      → GovernanceEnvelope store    ← SQLite WAL, append-only audit log
      → idempotency_keys store      ← inter-session exactly-once guarantee

Strict isolation: the gateway never touches execution logic. The kernel never touches persistence or policy. Each layer has a single responsibility and cannot cross the boundary.

---

Install

pip install nano-vm-mcp

For programs with llm steps:

pip install 'nano-vm-mcp[litellm]'

---

MCP Tools

Tool	Description
run_program	Execute a Program dict → returns trace_id, status, step count, cost
get_trace	Retrieve full Trace JSON by trace_id
list_programs	List saved programs (id, name, created_at)
get_program	Retrieve saved Program JSON by program_id
delete_program	Delete a program and all its traces

---

Quick Start

stdio — Claude Desktop / local MCP client

nano-vm-mcp --transport stdio

claude_desktop_config.json:

{
  "mcpServers": {
    "nano-vm-mcp": {
      "command": "nano-vm-mcp",
      "args": ["--transport", "stdio"]
    }
  }
}

SSE — VPS / remote clients

NANO_VM_MCP_API_KEY=your-secret-token nano-vm-mcp --transport sse --port 8080

MCP client URL: http://<host>:8080/sse
Auth header: Authorization: Bearer your-secret-token

Docker Compose

services:
  nano-vm-mcp:
    image: ghcr.io/ale007xd/nano-vm-mcp:latest
    ports:
      - "8080:8080"
    volumes:
      - ./data:/data
    environment:
      NANO_VM_MCP_DB: /data/nano_vm_mcp.db
      NANO_VM_MCP_PORT: 8080
      NANO_VM_MCP_API_KEY: your-secret-token
    command: ["nano-vm-mcp", "--transport", "sse"]

---

Use with Claude Code Dynamic Workflows

nano-vm-mcp works as a governed execution backend for Claude Code dynamic workflows. While Claude Code orchestrates subagents dynamically, nano-vm-mcp adds what native subagents lack: deterministic FSM execution, replayable traces, exactly-once semantics, and an append-only audit trail per workflow step.

Why pair them

	Claude Code Dynamic Workflows	+ nano-vm-mcp
Parallel subagents	✅	✅
Dynamic orchestration	✅	✅
Deterministic step execution	❌	✅
Replayable audit trail per step	❌	✅
Inter-session idempotency	❌	✅
GDPR tombstoning	❌	✅
Capability enforcement (double gate)	❌	✅

Use this combination when a workflow subagent must execute a governed process — payment pipeline, approval chain, compliance check — where correctness and auditability matter beyond the LLM layer.

Setup

Install and start the server:

pip install nano-vm-mcp
nano-vm-mcp --transport stdio

Add to your Claude Code MCP configuration (project-level .mcp.json or ~/.claude/claude_desktop_config.json):

{
  "mcpServers": {
    "nano-vm-mcp": {
      "command": "nano-vm-mcp",
      "args": ["--transport", "stdio"]
    }
  }
}

Example: governed payment step inside a workflow

A Claude Code subagent calls run_program to execute a payment pipeline with full governance:

# Claude Code subagent calls this tool directly
result = await session.call_tool(
    "run_program",
    {
        "program": {
            "name": "payment_pipeline",
            "steps": [
                {"id": "validate",  "type": "tool", "tool": "validate_amount"},
                {"id": "reserve",   "type": "tool", "tool": "reserve_funds"},
                {"id": "capture",   "type": "tool", "tool": "capture_payment"},
                {"id": "receipt",   "type": "tool", "tool": "send_receipt",
                 "is_terminal": True},
            ]
        },
        "idempotency_key": "order-abc-123",  # exactly-once across retries and restarts
    }
)
# Returns: trace_id, status, step count, cost
# Every step produces a GovernanceEnvelope in SQLite — tamper-evident, append-only

The FSM kernel controls all state transitions. The subagent cannot skip steps, reorder execution, or bypass capability checks — regardless of what the LLM decides at the orchestration layer.

Retrieve the audit trail

After execution, any agent or observer can retrieve the full trace:

trace = await session.call_tool("get_trace", {"trace_id": result["trace_id"]})
# Returns: per-step status, duration_ms, usage, state_snapshots

Traces persist across sessions in SQLite WAL. trace_id is UUID4-stable for OTel propagation.

---

Configuration

Copy .env.example to .env:

cp .env.example .env

Variable	Default	Description
NANO_VM_MCP_DB	nano_vm_mcp.db	SQLite WAL database path
NANO_VM_MCP_HOST	0.0.0.0	SSE bind host
NANO_VM_MCP_PORT	8080	SSE bind port
NANO_VM_MCP_API_KEY	(unset)	Bearer token for SSE auth. If unset, all requests are allowed (warning logged)
NANO_VM_MCP_LLM_MODEL	(unset)	LiteLLM model string for llm steps (e.g. openrouter/meta-llama/llama-3.3-70b-instruct:free)

---

Endpoints

Path	Auth	Description
GET /health	none	Liveness probe — always returns {"status": "ok"}
GET /sse	bearer	SSE transport entry point
POST /messages	bearer	MCP message endpoint

---

Example: Run a Workflow

Payment pipeline — no LLM

program = {
    "name": "payment_flow",
    "steps": [
        {"id": "reserve",  "type": "tool", "tool": "reserve_funds"},
        {"id": "capture",  "type": "tool", "tool": "capture_payment"},
        {"id": "receipt",  "type": "tool", "tool": "send_receipt"},
    ]
}

No LLM. The gateway still guarantees: deterministic ordering, replayable trace, exactly-once semantics, append-only audit trail.

Async suspend / resume

Return the sentinel "PENDING" from any tool to suspend execution:

async def wait_bank_transfer(**kwargs) -> str:
    await register_webhook(kwargs["order_id"])
    return "PENDING"   # FSM → SUSPENDED, cursor persisted

FSM lifecycle: RUNNING → SUSPENDED → RUNNING → SUCCESS

This enables: payment settlement, courier confirmation, approval workflows, webhook orchestration, human-in-the-loop.

Note: "PENDING" is a reserved FSM sentinel. Use "REQUIRES_ACTION", "AWAITING_3DS", or any other string for domain-specific states.

Through MCP (SSE)

import asyncio
from mcp import ClientSession
from mcp.client.sse import sse_client

program = {
    "name": "demo",
    "steps": [
        {"id": "step1", "type": "tool", "tool": "hello_tool"}
    ]
}

async def main():
    headers = {"Authorization": "Bearer your-secret-token"}
    async with sse_client("http://localhost:8080/sse", headers=headers) as (r, w):
        async with ClientSession(r, w) as session:
            await session.initialize()
            result = await session.call_tool(
                "run_program",
                {
                    "program": program,
                    "save_as": "demo",
                    "idempotency_key": "order-abc-123",   # inter-session exactly-once
                }
            )
            print(result.content[0].text)

asyncio.run(main())

---

Idempotency — Inter-session Exactly-Once

Pass idempotency_key to run_program to guarantee that a program executes at most once per key, even across process restarts:

# First call — executes normally, result cached as "success"
result = await session.call_tool("run_program", {
    "program": program,
    "idempotency_key": "payment-order-xyz-001",
})

# Second call with same key — returns cached result immediately, no re-execution
result = await session.call_tool("run_program", {
    "program": program,
    "idempotency_key": "payment-order-xyz-001",
})

Crash recovery: if the process crashes after program start but before completion (status=pending), the next call with the same key overwrites the pending entry and re-executes. Once the result is written as status=success, it is immutable for that key.

This closes the inter-session duplicate risk that exists when a process restarts after creating a payment but before confirming it.

---

Governance Layer

GovernanceEnvelope

Each successful execution step produces an immutable GovernanceEnvelope stored in the governance_envelopes table:

Field	Type	Description
execution_id	str	Session / trace identifier
step_id	int	Step index within the execution
policy_hash	str	SHA-256 of the active PolicySnapshot
canonical_snapshot_hash	str	Merkle/delta hash of CanonicalState at this step
payload	dict | list	Projected (sanitized) step output

Envelopes are written only on error=None — they form a tamper-evident, append-only audit trail of successful transitions only.

PolicySnapshot and CapabilityRef — in depth

PolicySnapshot is a frozen Pydantic model created once per session. It carries the set of allowed tool names (tool_capabilities) and is hashed (SHA-256) before execution starts. Every GovernanceEnvelope records this hash — so any post-hoc modification of the policy is detectable.

from nano_vm.contracts import PolicySnapshot, CapabilityRef

policy = PolicySnapshot(
    tool_capabilities={"reserve_funds", "capture_payment", "send_receipt"},
)
# policy.hash() → SHA-256 hex, stored in every GovernanceEnvelope.policy_hash

CapabilityRef wraps sensitive values as opaque tokens (vault://secret/<id>) rather than storing raw plaintext in CanonicalState. The token is resolved JIT during tool execution and never written to the audit log.

ref = CapabilityRef(ref_id="card-4242", value="4242424242424242")
# Stored in state as: vault://secret/card-4242
# GovernanceEnvelope.payload contains the token, not the card number

GDPR Tombstoning

On a GDPR erasure event (E_gdpr_erase):

• Target ref is tombstoned (is_tombstone=True)
• All subsequent projections return [REDACTED_TOMBSTONE]
• The canonical_snapshot_hash chain remains valid — forensic auditability is preserved
• The secret is permanently gone

vm.erase(ref_id="card-4242")
# All future get_trace calls → payload contains "[REDACTED_TOMBSTONE]" for that field
# Hash chain remains intact — the erasure itself is auditable

Execution traces

Every step also writes a TRACE projection to the execution_traces table — a sanitized snapshot of state visible to downstream observers (LLMs, dashboards) with sensitive values replaced by CapabilityRef tokens:

steps = store.get_trace_steps(execution_id="exec-abc-123")
# [
#   {"step_index": 0, "step_id": "validate", "projected_json": "...", "canonical_hash": "..."},
#   {"step_index": 1, "step_id": "reserve",  "projected_json": "...", "canonical_hash": "..."},
# ]

---

Determinism and LLM Steps

nano-vm-mcp provides two distinct guarantees:

State determinism — the FSM kernel guarantees execution order, no step skipping, and reproducible trace structure regardless of LLM output. The graph of transitions is fixed at program definition time. This is unconditional.

Semantic determinism — the text produced by an LLM step may differ across runs even at temperature=0.0. nano-vm does not guarantee semantic determinism and does not try to.

These are orthogonal concerns. The runtime enforces state determinism; you control semantic determinism through prompt engineering and allowed_outputs.

Constraining LLM output at the runtime level

allowed_outputs (v0.8.0) validates the model's raw output against an explicit enum before it enters the FSM context — no prompt engineering required for enforcement:

{
    "id": "classify",
    "type": "llm",
    "prompt": "Is this a valid refund request? Reply ONLY with: yes or no",
    "output_key": "decision",
    "allowed_outputs": ["yes", "no"],   # runtime enforcement — not a prompt hint
    "on_error": "skip",                 # output → "yes" (first element) on mismatch
}

If the model returns anything outside ["yes", "no"], the runtime handles it according to on_error — without propagating the invalid value to downstream steps or condition expressions.

Condition expressions are evaluated by the ASTEngine — a sandboxed interpreter with no access to Python builtins. LLM output can appear as a value being tested, never as the condition expression itself:

# ✅ Safe — LLM output is a value, ASTEngine evaluates the expression
{"condition": "'yes' in '$decision'"}

# ❌ Never do this — condition expression must not come from LLM output
{"condition": user_supplied_expression}

---

Performance

The FSM runtime introduces near-zero overhead. The bottleneck is always the LLM API or external I/O.

Sequential execution (single FSM instance): the FSM processes one step at a time per execution_id. This is a deliberate design choice — it makes traces deterministic and replayable.

Parallel execution across independent workflows: run multiple FSM instances with separate execution_id values. The SQLite WAL store handles concurrent writers without locking.

parallel step type: within a single FSM, asyncio.gather fans out independent sub-steps concurrently. Wall-clock time equals the slowest sub-step.

Benchmarks (v0.7.3, Mock adapter, QEMU/KVM · Intel Xeon E5-2697A v4 · 2 cores · Python 3.12)

Scenario	Mean TPS	p95
Refund pipeline (sequential)	2,300/s	0.66 ms
MCP store round-trip	3,000/s	0.42 ms
GovernanceEnvelope write	1,300/s	171 ms
Parallel throughput (asyncio.gather)	436/s	542 ms
Replay equivalence	1,300/s	1.30 ms
Long-horizon (30-step program)	30/s	3,606 ms

For high-throughput scenarios: fan out across multiple execution_id instances rather than serializing through a single FSM. Each instance is independent, lightweight, and SQLite WAL handles concurrent writes safely.

---

Security

Condition expressions — ASTEngine

run_program accepts a full Program dict including condition steps with expression strings. These are evaluated by the ASTEngine — a deterministic sandboxed interpreter with no access to Python builtins, attribute access, or callable invocation.

Supported operators: ==, !=, >, <, in, not in, and, or, not, contains, dotted-path $var.field.

eval() is not used anywhere in the production execution path.

Rules for safe use:

• Condition logic must be authored by you, not generated from untrusted input at runtime.
• LLM output may appear as a value being tested ('yes' in '$decision'), never as the condition expression itself.
• If you expose this server to untrusted clients, validate or allowlist condition expressions before passing them to run_program.

Capability enforcement — double gate

Tool execution passes through two independent enforcement layers:

Layer	Mechanism
GovernedToolExecutor	Verifies tool name against PolicySnapshot.tool_capabilities; raises CapabilityDeniedError on violation
ExecutionVM (kernel)	Rejects any tool name not registered in the tool registry with VMError

Neither gate can be bypassed by LLM output. A tool not listed in the policy is never silently executed.

Avoid registering destructive or privileged tools (filesystem writes, shell exec, database mutations) without an explicit access control layer in your tool implementation.

SSE transport and auth

Set NANO_VM_MCP_API_KEY to enable bearer token authentication. The comparison is timing-safe (secrets.compare_digest). If unset, a warning is logged and all requests are allowed — suitable for localhost only.

Do not expose the SSE endpoint to the public internet without NANO_VM_MCP_API_KEY set or behind a reverse proxy with auth (nginx, Cloudflare Access, VPN).

---

Observability

Every execution exposes:

trace.trace_id          # UUID4 — stable for OTel propagation
trace.status            # SUCCESS | FAILED | SUSPENDED | BUDGET_EXCEEDED | STALLED
trace.final_output
trace.steps             # per-step: step_id, status, duration_ms, usage
trace.state_snapshots   # list[(step_index, sha256_hex)]

Traces are persisted to SQLite and retrievable by trace_id across sessions via get_trace.

---

Execution State Model

CREATED
  ↓
RUNNING ──── tool returns "PENDING" ──→ SUSPENDED
  │                                          │
  │                                    resume_with_program()
  │                                          │
  └──────────────────────────────────────────┘
  │
  ├── no more steps ──→ SUCCESS
  ├── tool error (on_error=fail) ──→ FAILED
  ├── max_steps / max_tokens exceeded ──→ BUDGET_EXCEEDED
  └── max_stalled_steps exceeded ──→ STALLED

Terminal states: SUCCESS, FAILED, BUDGET_EXCEEDED, STALLED. All are immutable.

---

Relationship to nano-vm

Layer	Responsibility
llm-nano-vm (kernel)	Deterministic FSM execution, ASTEngine, ProjectionLayer, step lifecycle
nano-vm-mcp (gateway)	MCP transport, persistence, governance, idempotency, capability enforcement

The gateway never owns transition logic. The FSM kernel does.

The kernel is MIT-licensed, independently versioned on PyPI (llm-nano-vm), and fully documented. Either layer can be used standalone or replaced — the boundary between them is a stable Python int