OpenViSTAI · Agents · Blockchain · Crypto
Submit a project
Menu
Agent Security & Sandboxing

Coderunner

Coderunner is an open-source, self-hosted local sandbox for running AI agents and their actions. It uses a container-based environment and exposes tools such as Python code execution through an MCP server for supported assistant integrations.

Explore the practical guide ↓

Independent OpenViST project profile, based on linked sources.

Coderunner — OpenViST project profile in Agent Security & Sandboxing.
OpenViST project illustration. See the linked sources for product details.
Understand the project

Practical overview

When it fits

Use Coderunner when you want a locally hosted execution environment for AI-agent workflows and can meet its documented macOS Apple Silicon and Python requirements.

Who it is for

  • Developers who want to run Claude Code or multiple AI agents in a local sandbox.
  • Teams and individual builders connecting coding assistants to persistent or isolated Python execution through MCP.

Documented uses

  • Run an agent’s Python code in a sandbox instead of directly in the host environment.
  • Maintain separate named Python sessions for different agents, or resume a sandbox while preserving uploads, kernels and installed packages.
  • Connect supported assistants to the sandbox through MCP.
  • Choose a host-only network mode without internet access when creating the sandbox.
Before you choose

Requirements & limits

The project is self-hosted and the repository is public under the Apache-2.0 license. The quick-start instructions specify macOS on Apple Silicon (M1/M2/M3/M4) and Python 3.10 or later.

  • The documented quick start requires macOS on Apple Silicon and Python 3.10 or later.
  • Outbound network access is unrestricted by default; no-internet host-only mode must be selected at installation and cannot be changed when resuming the container.
  • Named Python sessions can isolate kernel state between agents, with up to five concurrent sessions documented.
  • The project documents multiple MCP integrations, each with its own configuration steps.

Check current requirements in the official documentation. Source-based guidance does not mean OpenViST has run or independently tested this project.

Source-based practical guide

Install Coderunner and choose a network mode

Set up the local sandbox with either its default network access or a host-only network without internet access.

Before you start

  • Mac running macOS on Apple Silicon (M1, M2, M3 or M4)
  • Python 3.10 or later

Steps

  1. Clone the repository with the command below and change into the coderunner directory.

    Documented command
    git clone https://github.com/instavm/coderunner.git

  2. Make the installer executable with chmod +x install.sh.

  3. Run ./install.sh for the default network mode, or run CODERUNNER_NETWORK=none ./install.sh to create a host-only sandbox without internet access.

Expected outcome

The installer creates the local sandbox. With network mode set to none, the MCP server is documented at `http://127.0.0.1:8222/mcp`.

Guide limitations

  • The documented quick start specifies macOS on Apple Silicon and Python 3.10 or later.
  • Network mode is fixed when the container is created; the installer refuses to resume it with a different mode.
Documentation 1 ↗
Source-based practical guide

Connect Claude Desktop to Coderunner’s MCP proxy

Configure Claude Desktop to use the local Coderunner MCP proxy.

Before you start

  • A Coderunner installation; its installer creates ~/.coderunner/venv for the Claude Desktop proxy.
  • The cloned Coderunner repository and Claude Desktop.

Steps

  1. In the repository, copy examples/claude_desktop/claude_desktop_config.example.json to examples/claude_desktop/claude_desktop_config.json.

  2. Replace the placeholder Python and repository paths with the full paths to ~/.coderunner/venv/bin/python and the cloned repository.

  3. In Claude Desktop, open Settings → Developer and add the MCP server configuration.

  4. Restart Claude Desktop.

Expected outcome

Claude Desktop is configured to use Coderunner as an MCP server and can send code-execution requests to the sandbox.

Guide limitations

  • The configuration must use the actual local paths for the Python executable and repository.
Documentation 1 ↗
Evidence and transparency

Sources & review

Official repository ↗
Source record details
Source availability
Active
Sources checked
12 October 2026
Original website title
instavm/coderunner · Official homepage title · retained in its source language
How we review sourcesSuggest a correction

Related options

InstaVM — The README identifies InstaVM as a cloud-managed VM option for AI agents, while Coderunner is a local sandbox.