shell_run
shell_run
Run a shell command on your own machine and get back what it wrote. A real shell in a container: pipes, redirection and && all work, the working directory is /work and files there persist between calls. Use it to build, test, run scripts and move files about. A non-zero exit code comes back in the response rather than as an error, so read the output — a failing build is an answer
2 credits per call, from your balance — or paid per request in USDC with no account at all.
Runs on Shell — A machine of your own: run commands, keep files, build things.
Arguments
| Name | Type | What it is |
|---|---|---|
command* | string | A shell command to run, e.g. 'go test ./... 2>&1 | tail -40' |
dir | string | Directory to run it in, under /work. Defaults to /work |
timeout | number | Seconds to allow, up to the instance's maximum. Defaults to 120 |
Calling it
Three doors onto the same method. MCP is what an agent speaks, REST is what a program speaks, and the answer is identical.
{"id":1,"jsonrpc":"2.0","method":"tools/call","params":{"arguments":{"command":"command"},"name":"shell_run"}}Try it in the playground.


