The Systeme.io MCP server does more than its own documentation admits. The last article in this series demonstrated digital products, offers and funnels working through it, none of which the documentation lists. That is the good news.
Bigger than documented is not the same as complete. This is the other half: what is genuinely still missing once you are connected, and where the gaps sit in the protocol rather than in the paperwork.
What's still missing, documented or not
Some gaps are not a documentation lag. Webhooks are a full create-read-update-delete resource on the REST API — you can register one, inspect it, change it, remove it. There is no MCP equivalent. An assistant connected over MCP can create the digital product but cannot wire up the notification when someone buys it.
The same pattern holds for anything that needs to happen in bulk. Every MCP tool operates on one record. Tagging four contacts is four tool calls; tagging four thousand is four thousand, each one counted against the same rate limit as everything else you are doing that hour. The REST API and the dashboard's own import tools remain the better route for a genuine bulk migration.
The confirmation step belongs to your client, not the protocol
OpenAI's own documentation is specific about this: write actions in ChatGPT "by default require confirmation." By default is the phrase to notice. That is a client setting, not a guarantee the MCP server makes.
The server itself has no concept of a confirmation step. It exposes tools that create, update and delete, including ones that overwrite a live funnel page's content directly. Whether a human reviews that write before it happens is entirely up to the assistant you are using and how it is configured that day.
Practically: do not assume the safety net travels with you if you switch clients, plans or a workspace setting. Confirm write confirmation is actually on before you hand over anything that touches a live page or a real customer.
You can't see your rate-limit budget until you're inside it
The MCP server follows the same rate-limiting policy as the public API, and that policy publishes the mechanism, not a number. The developer reference describes a refilling quota and three headers — the limit, the remaining balance, the refill timing — without ever stating what the limit actually is.
The quota is also shared across every key on the account, not issued per key. A heavy automated task run through one key draws down the same budget a different key needs for something else. You find out you were close to the ceiling when a request returns 429, not before.
Built for one record at a time
Pagination on the account's own MCP tools is cursor-based and capped, not a free-text export. It is the right shape for "find this one contact" or "list the tags on this record," and the wrong shape for pulling an entire list out in one pass.
That is a reasonable design for a conversational interface, but it means the honest test before depending on MCP for something is not "can it do this at all," it is "can it do this one record at a time, inside a shared rate limit, without a bulk operation to fall back on."
Where the boundary actually sits
Not available at all
- Webhooks — full CRUD on the REST API, no MCP equivalent
- Bulk or batch operations — every tool acts on one record
- A published, numeric rate-limit figure
Available, with a catch
- Direct page and funnel content writes — no server-side confirmation step
- Rate limits — real, shared across every key on the account, size undisclosed
- Cursor-based pagination — right for a lookup, wrong for an export
It changes without telling you
The MCP server documentation describes itself as an initial release, and states plainly that Systeme.io is "actively improving it." That has already shown up as a practical problem twice in the last two weeks of researching this series: the ChatGPT setup guide still says to authenticate with a key in the URL, and the live server now answers with a full OAuth challenge the guide never mentions.
Treat every specific in this article the same way. Re-check the tool list your own client shows you before a task that matters, rather than trusting a list — this one included — that was accurate on the day it was written.
Who this is safe for right now
- Safe now: one-off, high-judgment tasks on a small number of records, with write confirmation deliberately left on.
- Safe now: read-heavy work — looking up a contact, checking a funnel, drafting a newsletter before you send it yourself.
- Wait: anything that depends on a webhook firing, since there is nothing on the MCP side to build that on.
- Wait: real bulk work — a list migration, a mass tag cleanup, anything sized for the REST API or the dashboard's own import tools instead.
- Wait, or build a manual check: any workflow where a wrong write to a live page would be expensive to notice late.
Technical disclosure
The documented MCP surface, its initial-release status and its rate-limiting headers were read from Systeme.io's own MCP server documentation on 27 September 2026. The REST API's webhook CRUD support and its roughly 130 documented endpoints come from the developer documentation index, checked the same date. The absence of a published numeric rate limit was confirmed directly from the public API reference page, which describes the header mechanism without stating a figure. OpenAI's confirmation-by-default wording for write actions comes from its developer-mode documentation. The pagination cap and the direct page-write tools were observed from the MCP tool definitions exposed to this account's own connection on the same date, not from a vendor document — the account is the account this series has used throughout, and capability there can move before any of these pages update to describe it.
Audit your page free, then come back to the list.