SandboxRunner
# SandboxRunner ๐ณ
**SandboxRunner** is a custom [Model Context Protocol (MCP)](https://modelcontextprotocol.io/) server that enables AI assistants and MCP-compatible clients to securely execute **Python** and **C++** code snippets inside disposable, isolated Docker containers โ and receive back structured results (stdout, stderr, exit code, and timing) in real time.
> Built to understand MCP server design, Docker-based process isolation, and security-boundary thinking from the ground up.
---
## Table of Contents
- [Features](#features)
- [Architecture](#architecture)
- [Requirements](#requirements)
- [Installation](#installation)
- [Configuration](#configuration)
- [MCP Tools](#mcp-tools)
- [Isolation & Resource Limits](#isolation--resource-limits)
- [Execution History](#execution-history)
- [Usage](#usage)
- [Project Structure](#project-structure)
- [Testing](#testing)
- [Known Limitations](#known-limitations)
- [Contributing](#contributing)
---
## Features
- โ
**Python execution** โ runs snippets inside `python:3.12-slim`
- โ
**C++ execution** โ two-stage compile (`g++ -std=c++17 -O2`) + run, using `gcc:14`
- โ
**Strong isolation per run**:
- No network access (`--network none`)
- Hard memory cap (default `256m`)
- Hard CPU cap (default `0.5` cores)
- Read-only root filesystem with a temporary scratch-only mount
- `noexec` tmpfs for `/tmp`
- Containers are ephemeral โ deleted after every run
- โ
**Independent timeout enforcement** โ a hung snippet cannot hang the MCP server; the container is force-killed on timeout
- โ
**Output size capping** โ stdout/stderr truncated at 100 KB with a clear marker
- โ
**Distinct compile vs. runtime errors** for C++ โ compiler errors are surfaced separately from runtime crashes
- โ
**Local execution history** โ every run is logged to a local SQLite database and queryable via MCP
- โ
**Actionable Docker errors** โ clear error message if Docker is not running, instead of a silent hang
- โ
**All tunables in one place** โ `config.py` is the single source of truth for limits, images, and settings
---
## Architecture
```
MCP Client โโ(stdio / sse)โโโถ SandboxRunner (FastMCP)
โ
โผ
execution.py (orchestration)
โ
โโโโโโโโโโโโโโโดโโโโโโโโโโโโโโ
โผ โผ
Docker (Python) Docker (C++)
python:3.12-slim gcc:14 container
--network none --network none
--memory 256m --memory 256m
--cpus 0.5 --cpus 0.5
--read-only Stage 1: compile (rw /scratch)
Stage 2: run (ro /scratch)
โ
โผ
database.py โ sandbox_history.db (SQLite)
```
### Per-execution flow (`run_code`)
1. Validates input โ language, code size, timeout ceiling
2. Writes code to a temporary host directory
3. Mounts that directory into a fresh Docker container as `/scratch`
4. **Python**: runs `python /scratch/snippet.py` directly
5. **C++**: compiles to `/scratch/a.out` (read-write mount), then runs the binary (read-only mount) in a second container โ compile errors are returned distinctly from runtime errors
6. Enforces timeout at the server level; kills the container if exceeded
7. Captures and truncates stdout/stderr
8. Logs run metadata to SQLite
9. Returns a structured result to the MCP client
---
## Requirements
- **Python โฅ 3.12**
- **[uv](https://docs.astral.sh/uv/)** โ package manager (`pip install uv` or see [uv docs](https://docs.astral.sh/uv/getting-started/installation/))
- **Docker Desktop or Docker Engine** โ must be installed and running
- Docker images (pre-pull recommended):
- `python:3.12-slim`
- `gcc:14`
### Python Dependencies
| Package | Purpose |
|---|---|
| `mcp[cli] >= 1.28.1` | Official MCP SDK (FastMCP server + CLI) |
| `docker >= 7.1.0` | Docker SDK โ container orchestration |
**Dev only:**
| Package | Purpose |
|---|---|
| `pytest >= 9.1.1` | Test runner |
| `pytest-asyncio >= 1.4.0` | Async test support |
---
## Installation
```bash
# 1. Clone the repository
git clone https://github.com/huzayfaSiddique/sandbox_runner.git
cd sandbox_runner
# 2. Ensure Docker is running
docker ps
# 3. Install dependencies and create the virtual environment
uv sync
# 4. Pre-pull the Docker images (avoids slow cold start on first run)
docker pull python:3.12-slim
docker pull gcc:14
```
The `sandbox-runner` console script is installed automatically via `[project.scripts]` in `pyproject.toml`.
---
## Configuration
All tunables live in **`src/sandbox_runner/config.py`**. No environment variables are required beyond the optional transport flag.
| Setting | Default | Description |
|---|---|---|
| `DEFAULT_TIMEOUT_SECONDS` | `10` | Default execution timeout |
| `MAX_TIMEOUT_SECONDS` | `30` | Hard ceiling โ cannot be exceeded by callers |
| `MEMORY_LIMIT` | `"256m"` | Per-container memory cap |
| `CPU_LIMIT` | `"0.5"` | Per-container CPU share (in cores) |
| `MAX_CODE_SIZE_BYTES` | `51200` (50 KB) | Maximum allowed snippet size |
| `MAX_OUTPUT_SIZE_BYTES` | `102400` (100 KB) | Output truncation threshold |
| `DB_FILE` | `"sandbox_history.db"` | SQLite file for execution history |
| `DEFAULT_HISTORY_LIMIT` | `20` | Default rows returned by history tool |
To **add a new language**, add an entry to `SUPPORTED_LANGUAGES` in `config.py` with an `image`, `run_cmd`, and optionally `compile_cmd` / `source_file` for compiled languages.
---
## MCP Tools
| Tool | Description | Inputs | Outputs |
|---|---|---|---|
| `run_code` | Execute a code snippet in an isolated container | `language` (`python`\|`cpp`), `code`, `timeout_seconds` (opt, default `10`, max `30`) | `status`, `exit_code`, `stdout`, `stderr`, `duration_ms`, `language` |
| `list_supported_languages` | List available languages and Docker images | โ | `[{language, image, description}]` |
| `get_execution_history` | Retrieve recent run records | `limit` (opt, default `20`) | `[{id, timestamp, language, code_snippet, status, exit_code, duration_ms, stdout_size, stderr_size}]` |
**`status` values:** `success` ยท `error` ยท `timeout` ยท `compile_error` (C++ only)
**Example `run_code` response:**
```json
{
"status": "success",
"exit_code": 0,
"stdout": "The sum of elements is: 15\n",
"stderr": "",
"duration_ms": 1823.47,
"language": "cpp"
}
```
---
## Isolation & Resource Limits
Each run is sandboxed with the following Docker constraints:
| Constraint | Value | Effect |
|---|---|---|
| `--network none` | Enforced | No outbound or inbound network access |
| `--memory 256m` | Configurable | Hard memory ceiling per container |
| `--cpus 0.5` | Configurable | CPU share cap |
| `--read-only` | Always on | Root filesystem is immutable |
| `tmpfs /tmp` | `size=64m,noexec` | Small, non-executable temp space |
| `--rm` | Always on | Container is deleted after each run |
| Timeout kill | Server-level | Server kills container if it exceeds timeout |
> โ ๏ธ This uses Docker-level isolation โ suitable for personal/local use. It is **not** a hardened multi-tenant sandbox (e.g. gVisor, Firecracker) and is **not** intended for running untrusted third-party code.
---
## Execution History
Every run is stored in a local SQLite database (`sandbox_history.db`):
```sql
CREATE TABLE IF NOT EXISTS runs (
id INTEGER PRIMARY KEY AUTOINCREMENT,
timestamp REAL NOT NULL,
language TEXT NOT NULL,
code_snippet TEXT NOT NULL, -- first 500 chars only
status TEXT NOT NULL,
exit_code INTEGER,
duration_ms REAL NOT NULL,
stdout_size INTEGER NOT NULL DEFAULT 0,
stderr_size INTEGER NOT NULL DEFAULT 0
);
```
- Only the **first 500 characters** of each snippet are persisted as a preview.
- `stdout_size` / `stderr_size` store byte counts, not the full content.
- Timestamps are returned as **ISO 8601 UTC** strings via the `get_execution_history` tool.
---
## Usage
### Registering with an MCP Client
Add SandboxRunner to your MCP client's configuration. Example for **Claude Desktop** (`claude_desktop_config.json`):
```json
{
"mcpServers": {
"sandbox-runner": {
"command": "uv",
"args": [
"--directory",
"/absolute/path/to/sandbox-runner",
"run",
"sandbox-runner"
]
}
}
}
```
> Replace `/absolute/path/to/sandbox-runner` with your actual cloned directory path.
> On Windows: `C:\\Users\\YourName\\path\\to\\sandbox-runner`
Restart the MCP client after saving the config. The three tools will become available immediately.
### Running Standalone
```bash
# Start with stdio transport (default โ for MCP clients like Claude Desktop)
uv run sandbox-runner
# Start with SSE transport (for web-based or HTTP MCP clients)
uv run sandbox-runner --transport sse
# View CLI help
uv run sandbox-runner --help
```
---
## Project Structure
```
sandbox-runner/
โโโ src/
โ โโโ sandbox_runner/
โ โโโ __init__.py # Package version
โ โโโ config.py # All tunables โ limits, images, DB path
โ โโโ database.py # SQLite connection, record_run, fetch_history
โ โโโ execution.py # Docker orchestration, validation, truncation
โ โโโ main.py # CLI entrypoint (argparse + transport)
โ โโโ server.py # FastMCP server + tool definitions
โโโ tests/
โ โโโ test_execution.py # Input validation + mocked Docker execution tests
โ โโโ test_server.py # Mocked MCP tool-level tests
โโโ pyproject.toml # Project metadata, deps, build config
โโโ uv.lock # Locked dependency tree
โโโ .python-version # Pinned Python version
โโโ sandbox_history.db # SQLite history (auto-created at runtime)
โโโ README.md
```
---
## Testing
The test suite uses mocked Docker calls โ **Docker does not need to be running** to run the tests.
```bash
uv run pytest
```
**Current coverage:**
- `test_execution.py`
- Valid input passes without errors
- Unsupported language raises `ValueError`
- Empty code raises `ValueError`
- Oversized code raises `ValueError`
- Timeout exceeding max raises `ValueError`
- Mocked end-to-end Python execution โ asserts `status == "success"` and correct stdout
- `test_server.py`
- `run_code` success path โ verifies result format and DB logging call
- `run_code` invalid language โ verifies error response and that DB is **not** written
- `list_supported_languages` โ returns entries for both `python` and `cpp`
- `get_execution_history` โ timestamps formatted as ISO 8601 strings
---
## Known Limitations
- **Docker must be running.** The server returns a clear error if the Docker daemon is unreachable โ no silent hangs.
- **Docker-level isolation only.** Not a hardened cloud-grade sandbox. Designed for single-user, local use.
- **No multi-tenancy.** No auth, no rate limiting โ assumes a single trusted operator.
- **Fixed resource limits by default.** 256 MB / 0.5 CPU / 10s default timeout may be restrictive for heavy compute workloads.
- **History stores only snippet previews.** Full code content is not persisted; only the first 500 characters are stored.
---
## Contributing
Contributions are welcome!
1. Fork the repo and create a feature branch
2. Run `uv run pytest` and ensure all tests pass
3. Keep new tunables in `config.py` โ avoid hardcoding values elsewhere
4. Open a Pull Request with a clear description of the change and its motivation
---
TDQS
Scored across 3 tools
Each tool has a clearly distinct purpose: run_code executes code snippets, list_supported_languages returns available runtimes, and get_execution_history provides a log of past runs. There is no functional overlap.
All tool names follow the verb_noun pattern (run_code, list_supported_languages, get_execution_history) with consistent snake_case style.
Three tools is a well-scoped set for a code sandbox server. It covers the essential operations: running code, querying supported languages, and reviewing execution history, without unnecessary extras.
The tool surface is complete for the stated purpose of running code in an isolated sandbox. It includes language discovery, code execution, and history auditing, with no obvious missing functionality.