Koast’s MCP publishing limit is not a daily budget: keep the two separate before connecting

Koast authorizes reading, editing, and publishing separately. Its publishing limit caps the cumulative amount a single connection can launch, but cannot replace budget management in the advertising account.

Source: docs.koast.ai
November 16, 2026
3 min

For teams that want to review ad performance and prepare new ads in AI conversations, the most important aspect of Koast’s MCP integration is not the number of tools, but the permission boundaries between reading, editing, and publishing. One distinction is particularly important: a connection’s publishing limit is not the advertising account’s daily budget. This analysis is based on official documentation; no live advertising was carried out.

Read the data first, then decide whether to enable execution

Koast’s official connection guide says MCP is available on paid plans with an active subscription. Clients supporting remote MCP and OAuth can connect by signing in, without self-hosting a service. This brings advertising metrics and campaign preparation into the same conversation, reducing screenshots, data copying, and manual handoffs.

However, a successful connection does not mean every permission should be enabled. Koast separates Read, Write, and Publish. Colleagues who only prepare weekly reports can start with a read-only connection; media buyers who need to prepare ads can then consider editing access. The guide’s Publish and spend option is off by default and explicitly covers launching ads and changing live budgets. A broad request such as “help me optimize” should not be enough to turn a reporting assistant into an operator authorized to spend.

Publishing allowances and account budgets solve different problems

The permissions documentation defines the publish limit as the maximum cumulative amount a connection is allowed to launch over its entire lifetime, rather than a budget that resets daily. If no limit is set, publishing is denied and the relevant tools do not appear. This limit is independent of the advertising account’s budget.

For example, if a team plans to have an assistant help with a small-scale test, it should separately determine the amount this connection may commit, as well as the daily budget and campaign duration in the advertising account. Setting a number on the connection does not establish that actual spending across all accounts is capped by that same number; the official definition makes no such guarantee. This also means that managing connection replacements or revocations must be kept separate from checking existing ads. Disconnecting the assistant should not be treated as stopping ad delivery.

Use a limited acceptance test to decide whether to expand permissions

Official documentation stipulates that only one brand is operated on per conversation. The tools that are visible also depend on authorization, team role, plan, and the advertising provider’s supported capabilities. Agencies serving multiple clients should be particularly careful to confirm the current brand before checking results, so that objectives from another account are not carried into the current operation.

A suggested approach is to select one brand, ask the assistant to read metrics for a fixed period, and compare them with existing reports. Once the reporting definitions are confirmed, prepare drafts, have a person check the audience, creative, and budget, and only then decide whether to enable publishing. This is a recommended acceptance-testing method, not a personally tested workflow. If the need is simply to explain reports, read-only permissions already deliver the core benefit. If direct execution is needed, someone must also be responsible for approvals, checking what went live, and managing account budgets.

Content note:This article is based on publicly available information and is intended as a reference for evaluating tools. Features, pricing, and related policies may change at any time; please refer to the latest official information.

Original source:docs.koast.ai