Using log well

How we use it

We run log from Claude Code, through the MCP tools. We tell the agent what we want in plain words, and it follows the matching use case page step by step. You can type the same commands yourself in a terminal.

Start from a use case

Pick the use case closest to what you want and tell your agent:

go on docs/useCases/twitter/dmInbox.md

Each page says which tool runs each step and what shows it worked, so your agent doesn't improvise.

Nothing acts until you say so

Anything that posts, sends, likes or follows is a dry run by default. It shows what it would do and sends nothing.

log discord send --account yourname --channel "Weekend plans" --text "hello"          # dry run
log discord send --account yourname --channel "Weekend plans" --text "hello" --live --approvedBy you

Over MCP, the acting version is its own tool ending in Live (for example discordSendLive), and it needs confirmLive: true. Reading is always safe.

Each account paces itself

Every account keeps its own:

• daily limits on how much it writes
• pacing between actions, so nothing goes out back to back
• waking hours in its home timezone, outside which it won't run

If an account is refused, the message names an error code. log errors says what it means and what to do.

Scripts

Add --json to get machine-readable output. A command with --json never prompts.

log localAccount list --json

Every command and flag is in the reference.