{ "@context": "https://schema.org", "@id": "https://aggr.aymericbeaumet.com/items/philschmid-de/2025-07-22-code-sandbox-mcp-a-simple-code-interpreter-for-your-ai/#webpage", "@type": [ "WebPage", "ArchiveComponent" ], "articleBody": "July 22, 2025 4 minute read [View Code](https://github.com/philschmid/code-sandbox-mcp)\n\nCode agents are transforming software development. But how do we safely let them execute code? Code sandboxes have evolved from basic security tools into essential development infrastructure.\n\nToday, I'm launching **[Code Sandbox MCP](https://github.com/philschmid/code-sandbox-mcp)**, a lightweight, STDIO-based Model Context Protocol (MCP) Server, allowing AI assistants and LLM applications to safely execute code snippets using containerized environments. It is uses the [llm-sandbox](https://github.com/vndee/llm-sandbox) package for the containerization and execution of the code snippets.\n\nIt exposes `run_python_code` and `run_javascript_code` tools, giving your AI agent the ability to execute code on your own infrastructure.\n\n## How it works\n\n1. Starts a container session (podman, docker, etc.) and ensures the session is open.\n2. Writes the code to a temporary file on the host.\n3. Copies this temporary file into the container at the configured workdir.\n4. Executes the language-specific commands to run the code, e.g. python `python3 -u code.py` or javascript `node -u code.js`\n5. Captures the output and error streams from the container.\n6. Returns the output and error streams to the client.\n\n![Code Sandbox MCP](https://www.philschmid.de/static/blog/code-sandbox-mcp/code-sandbox-mcp.png)\n\nThis entire workflow happens on your machine, locally or on your own server, keeping your data and code private.\n\n## Get Started with the Gemini SDK\n\nIntegrating Code Sandbox MCP with your Python application using the Gemini SDK is incredibly straightforward. First, install the package:\n\nBash\n\n```bash\npip install git+https://github.com/philschmid/code-sandbox-mcp.git\n```\n\nThen, you can use `fastmcp` to connect your Gemini client to the local code execution tool:\n\nPython\n\n```python\nfrom fastmcp import Client\nfrom google import genai\nimport asyncio\n \n# Configure the MCP client to use the local server\nmcp_client = Client(\n {\n \"local_server\": {\n \"transport\": \"stdio\",\n \"command\": \"code-sandbox-mcp\",\n }\n }\n)\ngemini_client = genai.Client()\n \nasync def main():\n async with mcp_client:\n response = await gemini_client.aio.models.generate_content(\n model=\"gemini-1.5-flash\",\n contents=\"Use Python to ping google.com and return the response.\",\n config=genai.types.GenerateContentConfig(\n tools=[mcp_client.session], # Pass the FastMCP client session\n ),\n )\n # The model's response will include the output from the code execution\n print(response.text)\n \nif __name__ == \"__main__\":\n asyncio.run(main())\n \n```\n\n## Empower the Gemini CLI\n\nYou can also supercharge the Gemini CLI with code execution capabilities. Simply add the server configuration to your `~/.gemini/settings.json` file:\n\nJSON\n\n```json\n{\n \"mcpServers\": {\n \"code-sandbox\": {\n \"command\": \"code-sandbox-mcp\"\n }\n }\n}\n \n```\n\nNow, when you use the Gemini CLI, it can automatically discover and use the `run_python_code` tool to answer your questions!\n\n## Security\n\nBy leveraging `llm-sandbox`, Code Sandbox MCP benefits from multiple layers of security:\n\n1. **Container Isolation**: All code runs inside a sandboxed container (Docker, Podman, etc.), completely isolated from the host system's filesystem and processes.\n2. **Resource Limits**: The underlying sandbox can be configured with strict memory, CPU, and execution time limits to prevent resource exhaustion.\n3. **Network Controls**: You can define network policies to restrict or completely block outbound network access from the container.\n4. **Pre-execution Analysis**: The underlying `llm-sandbox` framework supports security policies that can analyze code for dangerous patterns (e.g., `os.system`, file system access) before it even runs.\n\n## Why Build This?\n\nI built Code Sandbox MCP primarily for my own use. When working with the Gemini CLI or other coding agents, I constantly struggled with testing code snippets in an isolated environment. Especially when you hand over more responsbility to the agent.\n\nWhile managed solutions like [Daytona](https://www.daytona.io/) and [E2B](https://e2b.dev/) are excellent for scaling or building customer facing agents, they didn't fit my specific needs. In my use cases, I wanted to give agents a way to execute code in an environment I can control and don't have to pay for.\n\nI mean:\n\n- **Having the libraries I need installed**: Unlike integrated code execution tools that come with limited packages, I can customize my environment with any dependencies.\n- **Access to secrets and credentials**: I can safely provide API keys or other sensitive information.\n- **Local code access**: Agents can work with my actual codebase, files, and local resources. I export files.\n- **No unnecessary costs**: There's simply no need to pay for a cloud solution when executing small code snippets for personal development work.\n\n* * *\n\nThanks for reading! If you have any questions or feedback, please let me know on [Twitter](https://twitter.com/_philschmid) or [LinkedIn](https://www.linkedin.com/in/philipp-schmid-a6a2bb196/).\n", "authors": [], "captured_at": "2026-09-27T19:29:15.293681927+00:00", "category": "programming", "content_html": "

July 22, 2025 4 minute read View Code

\n

Code agents are transforming software development. But how do we safely let them execute code? Code sandboxes have evolved from basic security tools into essential development infrastructure.

\n

Today, I'm launching Code Sandbox MCP, a lightweight, STDIO-based Model Context Protocol (MCP) Server, allowing AI assistants and LLM applications to safely execute code snippets using containerized environments. It is uses the llm-sandbox package for the containerization and execution of the code snippets.

\n

It exposes run_python_code and run_javascript_code tools, giving your AI agent the ability to execute code on your own infrastructure.

\n

How it works

\n
    \n
  1. Starts a container session (podman, docker, etc.) and ensures the session is open.
    2. \n
  2. Writes the code to a temporary file on the host.
    3. \n
  3. Copies this temporary file into the container at the configured workdir.
    4. \n
  4. Executes the language-specific commands to run the code, e.g. python python3 -u code.py or javascript node -u code.js
    5. \n
  5. Captures the output and error streams from the container.
    6. \n
  6. Returns the output and error streams to the client.
    7. \n
\n

\"Code

\n

This entire workflow happens on your machine, locally or on your own server, keeping your data and code private.

\n

Get Started with the Gemini SDK

\n

Integrating Code Sandbox MCP with your Python application using the Gemini SDK is incredibly straightforward. First, install the package:

\n

Bash

\n
pip install git+https://github.com/philschmid/code-sandbox-mcp.git\n
\n

Then, you can use fastmcp to connect your Gemini client to the local code execution tool:

\n

Python

\n
from fastmcp import Client\nfrom google import genai\nimport asyncio\n \n# Configure the MCP client to use the local server\nmcp_client = Client(\n    {\n        "local_server": {\n            "transport": "stdio",\n            "command": "code-sandbox-mcp",\n        }\n    }\n)\ngemini_client = genai.Client()\n \nasync def main():\n    async with mcp_client:\n        response = await gemini_client.aio.models.generate_content(\n            model="gemini-1.5-flash",\n            contents="Use Python to ping google.com and return the response.",\n            config=genai.types.GenerateContentConfig(\n                tools=[mcp_client.session],  # Pass the FastMCP client session\n            ),\n        )\n        # The model's response will include the output from the code execution\n        print(response.text)\n \nif __name__ == "__main__":\n    asyncio.run(main())\n \n
\n

Empower the Gemini CLI

\n

You can also supercharge the Gemini CLI with code execution capabilities. Simply add the server configuration to your ~/.gemini/settings.json file:

\n

JSON

\n
{\n  "mcpServers": {\n    "code-sandbox": {\n      "command": "code-sandbox-mcp"\n    }\n  }\n}\n \n
\n

Now, when you use the Gemini CLI, it can automatically discover and use the run_python_code tool to answer your questions!

\n

Security

\n

By leveraging llm-sandbox, Code Sandbox MCP benefits from multiple layers of security:

\n
    \n
  1. Container Isolation: All code runs inside a sandboxed container (Docker, Podman, etc.), completely isolated from the host system's filesystem and processes.
    2. \n
  2. Resource Limits: The underlying sandbox can be configured with strict memory, CPU, and execution time limits to prevent resource exhaustion.
    3. \n
  3. Network Controls: You can define network policies to restrict or completely block outbound network access from the container.
    4. \n
  4. Pre-execution Analysis: The underlying llm-sandbox framework supports security policies that can analyze code for dangerous patterns (e.g., os.system, file system access) before it even runs.
    5. \n
\n

Why Build This?

\n

I built Code Sandbox MCP primarily for my own use. When working with the Gemini CLI or other coding agents, I constantly struggled with testing code snippets in an isolated environment. Especially when you hand over more responsbility to the agent.

\n

While managed solutions like Daytona and E2B are excellent for scaling or building customer facing agents, they didn't fit my specific needs. In my use cases, I wanted to give agents a way to execute code in an environment I can control and don't have to pay for.

\n

I mean:

\n\n
\n

Thanks for reading! If you have any questions or feedback, please let me know on Twitter or LinkedIn.

\n", "content_markdown": "July 22, 2025 4 minute read [View Code](https://github.com/philschmid/code-sandbox-mcp)\n\nCode agents are transforming software development. But how do we safely let them execute code? Code sandboxes have evolved from basic security tools into essential development infrastructure.\n\nToday, I'm launching **[Code Sandbox MCP](https://github.com/philschmid/code-sandbox-mcp)**, a lightweight, STDIO-based Model Context Protocol (MCP) Server, allowing AI assistants and LLM applications to safely execute code snippets using containerized environments. It is uses the [llm-sandbox](https://github.com/vndee/llm-sandbox) package for the containerization and execution of the code snippets.\n\nIt exposes `run_python_code` and `run_javascript_code` tools, giving your AI agent the ability to execute code on your own infrastructure.\n\n## How it works\n\n1. Starts a container session (podman, docker, etc.) and ensures the session is open.\n2. Writes the code to a temporary file on the host.\n3. Copies this temporary file into the container at the configured workdir.\n4. Executes the language-specific commands to run the code, e.g. python `python3 -u code.py` or javascript `node -u code.js`\n5. Captures the output and error streams from the container.\n6. Returns the output and error streams to the client.\n\n![Code Sandbox MCP](https://www.philschmid.de/static/blog/code-sandbox-mcp/code-sandbox-mcp.png)\n\nThis entire workflow happens on your machine, locally or on your own server, keeping your data and code private.\n\n## Get Started with the Gemini SDK\n\nIntegrating Code Sandbox MCP with your Python application using the Gemini SDK is incredibly straightforward. First, install the package:\n\nBash\n\n```bash\npip install git+https://github.com/philschmid/code-sandbox-mcp.git\n```\n\nThen, you can use `fastmcp` to connect your Gemini client to the local code execution tool:\n\nPython\n\n```python\nfrom fastmcp import Client\nfrom google import genai\nimport asyncio\n \n# Configure the MCP client to use the local server\nmcp_client = Client(\n {\n \"local_server\": {\n \"transport\": \"stdio\",\n \"command\": \"code-sandbox-mcp\",\n }\n }\n)\ngemini_client = genai.Client()\n \nasync def main():\n async with mcp_client:\n response = await gemini_client.aio.models.generate_content(\n model=\"gemini-1.5-flash\",\n contents=\"Use Python to ping google.com and return the response.\",\n config=genai.types.GenerateContentConfig(\n tools=[mcp_client.session], # Pass the FastMCP client session\n ),\n )\n # The model's response will include the output from the code execution\n print(response.text)\n \nif __name__ == \"__main__\":\n asyncio.run(main())\n \n```\n\n## Empower the Gemini CLI\n\nYou can also supercharge the Gemini CLI with code execution capabilities. Simply add the server configuration to your `~/.gemini/settings.json` file:\n\nJSON\n\n```json\n{\n \"mcpServers\": {\n \"code-sandbox\": {\n \"command\": \"code-sandbox-mcp\"\n }\n }\n}\n \n```\n\nNow, when you use the Gemini CLI, it can automatically discover and use the `run_python_code` tool to answer your questions!\n\n## Security\n\nBy leveraging `llm-sandbox`, Code Sandbox MCP benefits from multiple layers of security:\n\n1. **Container Isolation**: All code runs inside a sandboxed container (Docker, Podman, etc.), completely isolated from the host system's filesystem and processes.\n2. **Resource Limits**: The underlying sandbox can be configured with strict memory, CPU, and execution time limits to prevent resource exhaustion.\n3. **Network Controls**: You can define network policies to restrict or completely block outbound network access from the container.\n4. **Pre-execution Analysis**: The underlying `llm-sandbox` framework supports security policies that can analyze code for dangerous patterns (e.g., `os.system`, file system access) before it even runs.\n\n## Why Build This?\n\nI built Code Sandbox MCP primarily for my own use. When working with the Gemini CLI or other coding agents, I constantly struggled with testing code snippets in an isolated environment. Especially when you hand over more responsbility to the agent.\n\nWhile managed solutions like [Daytona](https://www.daytona.io/) and [E2B](https://e2b.dev/) are excellent for scaling or building customer facing agents, they didn't fit my specific needs. In my use cases, I wanted to give agents a way to execute code in an environment I can control and don't have to pay for.\n\nI mean:\n\n- **Having the libraries I need installed**: Unlike integrated code execution tools that come with limited packages, I can customize my environment with any dependencies.\n- **Access to secrets and credentials**: I can safely provide API keys or other sensitive information.\n- **Local code access**: Agents can work with my actual codebase, files, and local resources. I export files.\n- **No unnecessary costs**: There's simply no need to pay for a cloud solution when executing small code snippets for personal development work.\n\n* * *\n\nThanks for reading! If you have any questions or feedback, please let me know on [Twitter](https://twitter.com/_philschmid) or [LinkedIn](https://www.linkedin.com/in/philipp-schmid-a6a2bb196/).\n", "dateCreated": "2026-09-27T19:29:15.293681927+00:00", "date_created": "2025-07-22T00:00:00+00:00", "date_modified": "2025-07-22T00:00:00+00:00", "date_published": "2025-07-22T00:00:00+00:00", "description": "Code Sandbox MCP is a simple, self-hosted code interpreter for your AI agents. It allows you to execute code snippets in containerized environments.", "external_url": "https://www.philschmid.de/code-sandbox-mcp", "git": { "commit": "ebb0941c5c06a2ab2377342d607535262a3e969e", "path": "items/philschmid-de/2025/07/2025-07-22-code-sandbox-mcp-a-simple-code-interpreter-for-your-ai.md", "permalink": "https://github.com/aymericbeaumet/aggr-instance/blob/ebb0941c5c06a2ab2377342d607535262a3e969e/items/philschmid-de/2025/07/2025-07-22-code-sandbox-mcp-a-simple-code-interpreter-for-your-ai.md" }, "id": "urn:aggr:item:9d58da1a47535b20a07bb33b6ad139c77f97f156", "isBasedOn": { "@id": "https://www.philschmid.de/code-sandbox-mcp" }, "mainEntity": { "@id": "https://www.philschmid.de/code-sandbox-mcp", "@type": "CreativeWork", "archivedAt": { "@id": "https://aggr.aymericbeaumet.com/items/philschmid-de/2025-07-22-code-sandbox-mcp-a-simple-code-interpreter-for-your-ai/#webpage" }, "dateModified": "2025-07-22T00:00:00+00:00", "datePublished": "2025-07-22T00:00:00+00:00", "headline": "Code Sandbox MCP: A Simple Code Interpreter for Your AI Agents", "timeRequired": "PT3M", "url": "https://www.philschmid.de/code-sandbox-mcp", "wordCount": 667 }, "name": "Archived snapshot: Code Sandbox MCP: A Simple Code Interpreter for Your AI Agents", "preview": null, "reading_minutes": 3, "replicated_at": null, "source": "philschmid-de", "source_name": "philschmid.de - RSS feed", "summary": "Code Sandbox MCP is a simple, self-hosted code interpreter for your AI agents. It allows you to execute code snippets in containerized environments.", "tags": [], "temporalCoverage": "2026-09-27T19:29:15.293681927+00:00", "title": "Code Sandbox MCP: A Simple Code Interpreter for Your AI Agents", "url": "https://aggr.aymericbeaumet.com/items/philschmid-de/2025-07-22-code-sandbox-mcp-a-simple-code-interpreter-for-your-ai/", "word_count": 667 }