# Run a Whole Brand From Your Terminal: How a Week Could Go

> Run a whole brand from your terminal with an MCP agent. A walkthrough of how a week could go, and the scopes and credit cap that keep it bounded.

Canonical page: https://www.socialfaktory.com/blog/run-a-whole-brand-from-your-terminal
Published: 2026-09-15
Language: en

You can run a whole brand from your terminal once an agent has an MCP endpoint and a token, and the interesting question is not whether that works. It is what a week of it would ask of you. This is a walkthrough, not a case study: how a week could go, the decisions you would make at each step, and the limits that keep it bounded.

Picture a brand with a few connected channels and a presenter already set up in the app. Before Monday you would mint one token for it, pin that token to the brand, and choose a monthly credit cap you would be comfortable losing in a bad week.

## What would you set up before Monday?

Three decisions, all in your settings. The scopes decide what the token may do: read is always on, generate lets it spend credits, and publish lets it compose, schedule, send and delete posts. The pin decides which brand's channels, generations and posts it can reach, while the credit wallet and uploads stay shared across the account. The cap decides how much it may start spending in a month.

A cautious first week leaves publish off, so the agent stops at generated video and text, because creating a post needs publish even for a draft. You can mint a token that carries it later, once the output stops surprising you.

Then comes the one integration step.

```bash
claude mcp add -s user --transport http socialfaktory https://www.socialfaktory.com/mcp --header "Authorization: Bearer sfp_..."
```

Everything after that command is plain English.

## What would Monday look like?

Mostly reading, and reading is free. Ask the agent to list the brand, its channels, the formats and the wallet balance, and none of those calls spends a credit. Then ask it to propose the week and price each video with the quote tool before it runs.

The quote is the useful artefact. A column of prices next to a column of ideas makes it easier to see which ideas are worth their cost, and cutting a weak idea at this stage costs nothing at all.

## What would the agent generate?

Whatever you approve from the quotes. A video starts with create_generation, which hands back an identifier instead of blocking, so a session can start several renders and read each one back when it is ready. Posts around each video come from generate_text. With the publish scope, create_posts then composes one post per channel, as drafts unless you ask for them to be scheduled.

That split between starting work and reading it back is what keeps a terminal session comfortable. You are never watching a spinner; you ask, move on, and come back.

## Where could it go wrong?

In the brief. If the agent misreads what you meant, it may generate something you did not want, and a finished render has used its credits whether or not you like the result. A render that fails on our side is refunded in full, and nothing comes back when the brief was wrong.

That is the risk to plan for. Quotes make the cost of a misunderstanding visible before it happens, and a specific brief is what prevents it. Name the product, the day and the angle. A two-line brief does more than three paragraphs of tone guidance, because the brand already holds the voice, the cast and the products.

## What changes when you add the publish scope?

The agent can then compose drafts and send posts, and you decide how sending works. send_post takes a mode: schedule puts the post on its channel at its planned time, and now publishes it immediately. A token that carries publish can do either without asking you again, so decide before you grant it whether you want to see the calendar first.

A token without publish can never post, whatever a session tries. That is the harder guarantee, and it is why a first week without publish is a sensible default rather than timidity.

## What would the daily rhythm be?

It could be short. A brief in the morning, a look at what it generated or scheduled afterwards, and the rest of the day spent elsewhere. Once the week has been quoted and approved on Monday, the remaining days are closer to review than to production.

How long that review takes depends on how often the drafts surprise you. The more precise the briefs, the less there is to change, and the shorter each morning becomes.

## What stops a week from running away?

The monthly credit cap. Any call that would take the token's spending past the cap is refused, and a session cannot talk its way around that. Raising the cap is a deliberate change in your settings, not something an agent can do for itself.

A cap is not a budget alarm; it is a hard limit. Together with the brand pin and the scopes, it is what makes leaving an agent running a reasonable thing to consider. The [agent hub](/agents) explains the three scopes and the cap together, because they only make sense as a set.

## What would you change in week two?

You would probably write a standing brief: a channel mix, a cadence and a boundary such as always quoting first. You would probably also settle the publish question one way or the other, based on how the first week's drafts read.

You might even write a deliberately vague brief for one piece, to see where ambiguity costs credits. With a low cap it is a cheap lesson, and it carries over to every brief you write after it.

## Is the terminal the right place to run a brand?

If you already work there, it removes a context switch, and the context switch is often the expensive part of posting consistently. If you would rather see previews while you decide, the [AI assistant](/features/ai-assistant) runs the same loop inside the app with the prices on the same screen. For the terminal, the [Codex guide](/agents/codex) and the [Claude guide](/agents/claude) have the setup, and the [MCP reference](/docs/mcp) lists every tool. [Start a brand](/register) and set the first cap low.

**In one sentence:** Run a whole brand from your terminal and the scopes, the credit cap and the idempotency keys do most of the work of keeping the week safe, because the agent can only do what its token allows.

## Questions

### Is this a record of a real customer week?

No. It is a walkthrough of how a week could run, not a report of one that happened.

### Can an agent publish a post immediately?

Yes, if its token carries the publish scope. send_post can schedule a post or publish it immediately.

### What stops an agent from overspending?

The monthly credit cap on the token. A call that would take spending past it is refused.

### Are credits refunded for a misunderstood brief?

No. A render that fails on our side is refunded in full, and one refused for content policy or a bad reference link keeps what it already spent.

### Can one session work across several brands?

Not with a pinned token. Pin one token per brand and each session stays on its own brand.

### What can an agent not do?

Clone a video from a link, make still images or carousels, or connect a channel.

## More detail

- All articles: https://www.socialfaktory.com/blog.md
- What SocialFaktory does: https://www.socialfaktory.com/features.md
- Pricing: https://www.socialfaktory.com/plans.md