The surface is deliberately tiny — a single call where the language is just a string — which means no per-language tool clutter in the context and no local toolchain to install for Rust, Swift or Scala. The cost model follows: you are handing execution to a hosted service under an API token, so treat the code you send as leaving the machine.
An MCP server in front of NetMind's hosted code execution service. The assistant sends source files and gets execution results back; the container is isolated, resource-limited and thrown away afterwards, so nothing runs on your machine.
- `execute_code` — pick a language, pass a list of files each with a name and full contents, and optionally stdin and command-line arguments
- Language identifiers covering python, javascript, typescript, java, cpp, c, go, rust, php, ruby, swift, kotlin, scala and r
- Multi-file programs, not just snippets — you send the whole set and it runs them together
- Results as stdout, stderr and data outputs on success, or an error description on failure
- Plots and charts your code writes are captured and returned as a link, with no extra step
- A sandboxed filesystem the program can read from and write to during the run
A NetMind API key, set as NETMIND_API_TOKEN. The documented client config runs it with uvx, so you need uv installed — via Homebrew or the install script. Nothing else is needed locally: execution happens on NetMind's infrastructure. MIT licensed.