Connect an agent

nox has an MCP server, so your agent can write scripts for you, test them, read the logs, fix what fails and schedule them, all from a chat. You approve the connection once, in nox, and can disconnect it whenever you like.

Your MCP server address

https://pernox.dev/mcp

Add nox to your agent

In your agent's settings, add a remote MCP server (sometimes called a connector), name it nox, and paste the address above. It uses the HTTP transport.
Leave any OAuth client ID and secret fields empty; nox registers the agent itself.
Connect. nox opens, asks you to log in, and shows what the agent will be able to do. Choose Allow.
Turn nox on in your agent's tools, then ask for what you need.

An agent that runs in the cloud connects from its provider's servers, so this nox has to be reachable on the public internet over HTTPS. An agent running on your own machine can connect to a local nox too.

Things to ask

"Write a script that fetches the Bank of England base rate every weekday at 07:00 London time and prints it. Test it first."
"My FX rates script failed last night. Look at the run and fix it."
"Pull yesterday's closing prices for these tickers and post a summary to my Google Sheet. I'll give you the API key."
"Which of my scripts haven't succeeded this week?"
"Move all my morning scripts to 06:30."

Your agent creates scripts paused, runs them to check the output, and only turns a schedule on when you say so.

Tools

What the nox MCP server lets your agent do. Most clients ask before each action.

list_scripts read-only

List the user's scripts with their schedules, next run times and the status of their last run.
get_script read-only

Get a script's code, packages and settings, the names of its keys (never their values), its next runs and its most recent runs.

script_id integer, required
The script's id, from list_scripts or create_script.
create_script

Create a Python script. It's created paused unless enabled is true; test it with run_script before turning the schedule on.

name string, required
A short name the user will recognise, e.g. "Gilt curve to sheet".

code string, required
The Python source. Print results to stdout; read secrets from os.environ.

requirements string
Packages to install, one per line in requirements.txt format, e.g. "requests\npandas==2.2.3".

schedule string
Five-field cron expression in the script's timezone. Defaults to "0 7 * * 1-5" (07:00 on weekdays).

timezone string
IANA timezone for the schedule and the script's TZ, e.g. "Europe/London". Defaults to UTC.

timeout_seconds integer
Kill the run after this many seconds (1-3600). Defaults to 300.

enabled boolean
Run on the schedule. Defaults to false; only set true once the user agrees.
update_script

Change any of a script's code, packages, name, schedule, timezone, timeout or whether it runs on schedule. Only the fields given are changed.

script_id integer, required
The script's id, from list_scripts or create_script.

name string
New name.

code string
New Python source, replacing the old.

requirements string
New packages, replacing the old list. An empty string removes them all.

schedule string
New five-field cron expression.

timezone string
New IANA timezone.

timeout_seconds integer
New timeout in seconds (1-3600).

enabled boolean
true to run on schedule, false to pause.
run_script

Run a script immediately and wait for it to finish, returning its status, exit code and logs. Waits up to wait_seconds (default 30, max 50); if it's still running, call get_run with the run_id. If the script is already running, waits on that run instead of starting another.

script_id integer, required
The script's id, from list_scripts or create_script.

wait_seconds integer
How long to wait for the run to finish (0-50). 0 returns straight away.
get_run read-only

Get a run's status, exit code, timings and logs, optionally waiting for it to finish.

run_id integer, required
The run's id, from run_script or list_runs.

wait_seconds integer
Wait up to this long for the run to finish (0-50). Defaults to 0.
list_runs read-only

List a script's recent runs, newest first, with status, trigger and duration. Use get_run for a run's logs.

script_id integer, required
The script's id, from list_scripts or create_script.

limit integer
How many runs to return (1-50). Defaults to 10.
set_key

Set an environment variable for a script, such as an API key or database URL. Values can't be read back. Only use values the user has given you for this purpose.

script_id integer, required
The script's id, from list_scripts or create_script.

name string, required
Variable name, e.g. DB_PASSWORD. Letters, numbers and underscores.

value string, required
The value.
delete_key destructive

Remove one of a script's environment variables.

script_id integer, required
The script's id, from list_scripts or create_script.

name string, required
The variable to remove.
delete_script destructive

Permanently delete a script with its keys and run history. Confirm with the user first.

script_id integer, required
The script's id, from list_scripts or create_script.

How scripts run

Language: Python, one file. Packages go in requirements.txt format, one per line.
Output: Anything printed to stdout or stderr is the run's log, up to 1 MB.
Success: Exit code 0 succeeds. A non-zero exit or an uncaught exception fails the run.
Keys: Environment variables, read with os.environ["NAME"]. Values are never shown again, to you or to your agent.
Time: TZ is set to the script's timezone, so datetime.now() is local time.
Schedules: Five-field cron, e.g. 0 7 * * 1-5 is 07:00 on weekdays. At most once every 1m0s.
Timeout: Up to 3600 seconds, then the run is stopped.
Files: The working directory is temporary and wiped after each run. Send results somewhere: an API, a database, a sheet.
Overlaps: A script never runs twice at once. A slot that comes up while it's still running is skipped.