11 tools · 12 languages · MCP-native

Code execution for any MCP-compatible AI agent

SandboxAPI ships an MCP server at https://mcp.sandboxapi.dev/mcp. Drop the JSON config below into Claude Desktop, Cursor, VS Code, or any other MCP-compatible client. Your AI can now run code in 12 languages, install packages, and keep state across calls.

11 tools, all the surface area you need

Every endpoint in the SandboxAPI HTTP API has a matching MCP tool. The agent can pick the mode that fits the moment.

execute_codeOne-shot synchronous execution

execute_batchRun up to 200 snippets in a single call

list_languagesDiscover available runtimes & versions

create_sessionOpen a stateful sandbox

session_executeExecute inside an open session

session_closeTear down a session early

session_install_packagespip / npm / gem / cargo

execute_asyncSubmit long jobs, get a token

get_executionPoll an async job's status

execute_with_expectedAuto-grade against expected output

get_capabilitiesRuntime feature discovery

Set up in 30 seconds

Pick your client, paste the JSON, replace YOUR_API_KEY with your RapidAPI or direct API key. Done.

Claude Desktop macOS / Win

Edit the Claude Desktop config and restart the app.

{
  "mcpServers": {
    "sandboxapi": {
      "url": "https://mcp.sandboxapi.dev/mcp",
      "headers": {
        "Authorization": "Bearer YOUR_API_KEY"
      }
    }
  }
}

~/Library/Application Support/Claude/claude_desktop_config.json

Cursor All platforms

Add to your Cursor MCP config (Settings → MCP).

{
  "mcpServers": {
    "sandboxapi": {
      "url": "https://mcp.sandboxapi.dev/mcp",
      "headers": {
        "Authorization": "Bearer YOUR_API_KEY"
      }
    }
  }
}

Settings → MCP → Add server

VS Code Continue / Cline

Compatible with the Continue and Cline MCP integrations.

{
  "mcpServers": {
    "sandboxapi": {
      "url": "https://mcp.sandboxapi.dev/mcp",
      "headers": {
        "Authorization": "Bearer YOUR_API_KEY"
      }
    }
  }
}

~/.continue/config.json or via the extension UI

What you can build

The MCP integration is the fastest way to ship one of these primitives in your AI agent product.

Code interpreter

Give your AI a persistent Python sandbox. Variables and DataFrames live across turns. Files persist. The agent can pip install what it needs and keep working.

create_session, session_install_packages, session_execute

Autonomous debugger

Failing test? The agent runs the code, sees the traceback, edits the file, runs again. Tight loops via stateful sessions instead of restart-and-pray.

create_session, session_execute, get_capabilities

REPL agent

Build the next "ChatGPT advanced data analysis" but on your own stack and your own model. 12 languages, gVisor isolation, and you're out in days.

create_session, session_execute, session_install_packages