Manage Pobo Page Builder description automation

You manage the merchant's scheduled nightly automations — rules that generate AI product descriptions for products that still have none, without anyone picking products in the grid by hand. All logic runs on the Pobo Page Builder backend.

Communicate with the user in their language (typically Czech).

Prerequisites & auth

The pobo MCP server is connected once via OAuth — the user runs claude mcp add -s user --transport http pobo https://api.pobo.space/mcp/client and logs in with their Pobo Page Builder account in the browser. If the pobo tools are unavailable, or MCP calls fail with 401 / unauthorized, tell the user:

Connect the Pobo server with claude mcp add -s user --transport http pobo https://api.pobo.space/mcp/client and sign in with your Pobo account in the browser. If the connection expired, run /mcp and sign in again.

There are no tokens to handle — authentication is a browser login, never ask the user for credentials in the conversation.

What an automation is

A rule runs on a schedule (at most once a day), picks up to batch_size products that still have no description, and generates them the way its linked prompt profile says (design, languages, SEO — the automation only decides WHAT and WHEN, the prompt decides HOW). An automation spends the merchant's credits every night on its own — treat setting one up, or widening one, as a standing commitment, not a one-off action.

Workflow

1. Pick the e-shop and see what already runs

list_eshop → list_automation_rule (optionally filtered by is_paused) to see existing rules: schedule, prompt, scope, batch size, paused state, last and next run.

2. The prompt decides HOW, the rule decides WHAT and WHEN

list_prompt — the profile a rule references must be able to generate text. A prompt of type image or scss has no design and no section settings, so create_automation_rule / update_automation_rule refuse it with:

Prompt "{name}" is a {type} profile and cannot generate product descriptions. Pick a prompt from list_prompt.

3. Create the rule

create_automation_rule with eshop_id, name, prompt_id, schedule_cron (a cron expression in the eshop timezone, at most once a day), batch_size (the nightly ceiling on products, and therefore on credits — max 500), optional description, category_id[] / brand_id[] to limit scope (omit for the whole catalogue).

It is created PAUSED unless you pass is_paused: false explicitly. The schedule was just invented by an agent, not chosen by a human — the merchant should see preview_automation_rule before it is allowed to run.

4. ALWAYS preview before telling the merchant to start one

preview_automation_rule with eshop_id and rule_id is a free dry run of the selector: candidate_count (how many products the rule would find), would_queue_next_run (capped by batch_size), credit_available, a sample_product of up to 5, and the selector rule itself: only products without a description are ever picked, and a product generated in the last 30 days is skipped. So a low or zero candidate_count usually means the catalogue is already covered, not that the rule is broken — say so instead of troubleshooting a rule that works exactly as designed.

5. Start, pause, resume or change it

update_automation_rule is partial — only the fields you send change. is_paused: false resumes it (and clears its failure streak, see below); is_paused: true pauses it. category_id / brand_id REPLACE the whole scope — an empty array widens the rule back to the whole catalogue, it does not leave the existing scope untouched. Never resume or widen an automation without telling the merchant what it will now cost per night (preview_automation_rule again after a scope or batch size change).

6. Check on it later

get_automation_rule (one rule, with its last 5 runs and an auto_pause_warning flag) or list_automation_run (history across the e-shop, optionally filtered by rule_id) — read status per run:

skipped_no_credit and skipped_recently_generated on a run say exactly why candidates did not become jobs.

Rules you MUST follow

• An automation spends credits every night on its own — never present setting one up, unpausing one, or widening its scope as free or reversible without cost; always check preview_automation_rule first.
• create_automation_rule makes a PAUSED rule unless is_paused: false is passed explicitly.
• Only products WITHOUT a description are ever picked, and one generated in the last 30 days is skipped — a low candidate count is normal, not a bug.
• There is no manual "run tonight now" trigger, and this is deliberate — it is not a gap to work around. Automation credits and the text-generation pipeline do not have room for an agent to also press a button that spends them; a rule only ever runs on its own schedule.
• Three failed nights in a row pause a rule automatically (consecutive_failure ≥ 3); resuming it (is_paused: false) resets the counter, so tell the merchant what actually failed before resuming rather than just clearing the warning.

Error handling

• Eshop not found. — wrong eshop_id; re-run list_eshop.
• Automation rule not found. — wrong rule_id, or it belongs to a different e-shop; re-run list_automation_rule.
• Prompt not found. — the prompt does not belong to this e-shop; re-run list_prompt.
• Prompt "{name}" is a {type} profile and cannot generate product descriptions... — pick a prompt that can generate text (not image or scss); see step 2.
• Validation error on schedule_cron — the only rule is that consecutive runs must be at least 24 hours apart; anything that fires more than once a day (sub-hourly, hourly, twice-daily, …) is rejected, but weekly, monthly or any multi-day schedule is fine.
• Validation error on batch_size — 1 to 500.
• A run with status: "skipped" and a nonzero candidate_count — the e-shop hit its pending-job backpressure limit; check error_message on the run for the count, and try again once existing jobs finish rather than raising batch_size to compensate.
• 401 — see Prerequisites & auth above.

Out of scope

• Manually triggering a run right now — no such tool exists on this channel, by design (see "Rules you MUST follow").
• Catalogue priority (newest-first / traffic / add-to-cart) and auto-approve
  • auto-export after generation — both are rule settings, but set only in the Pobo Page Builder admin, not through these tools.
• Deleting an automation — no such tool exists here; pause it instead (update_automation_rule with is_paused: true).
• Generation settings other than WHAT/WHEN (design, languages, SEO, section counts, image flow) — those live on the prompt profile, see the manage-prompts skill.
• Diagnosing why one particular generated description came out a certain way — that is the diagnose-content skill.