Two marketing platforms shipped an AI connector for their product in the same week. One quietly answered the governance question everyone else is skipping. The other did not even ask it.
The case for shipping fast
Start with the strongest version of the argument for moving quickly, because it is a real one. Every marketing operations team has a backlog of connector requests, and the Model Context Protocol exists precisely to remove the custom integration work that used to make each one a multi-week project. Eventtia, the event management platform, launched a native MCP server on September 25 that lets any MCP-compatible assistant, Claude, ChatGPT, Gemini, Copilot or Cursor, read and write events, attendees, sessions, speakers, check-in and payment data. It took the company, by its own account, under five minutes to connect, requires no code, and ships free with every plan. “Automating repetitive event tasks saves our clients hundreds of hours every month,” said Mauricio Palacio, CEO and co-founder of Eventtia. For a large share of what event operators actually do, drafting a check-in message, reconciling an attendee list, that speed is a straightforward win, and demanding a governance review before shipping it would be its own kind of failure.
Why the speed is the risk, not the feature
Here is the problem with stopping the argument there: read and write access to payments and attendee data is not a low-stakes connector, and Eventtia’s own announcement does not describe a permission model narrower than full account access. Compare it with what Stravito, a market-research platform, shipped one day earlier. Stravito’s MCP server, announced September 24, connects a company’s research library to the same set of assistants, but it runs read-only, inherits each requesting user’s existing permissions rather than granting the assistant its own, and cites its sources in every response. “The most valuable consumer research is worthless if it never actually guides real business decisions,” said Thor Olof Philogene, founder and CEO of Stravito, making the case for connecting research to where decisions get made. The company built the connection to answer that need without also building a new way for data to leak or a new account for an attacker to compromise.
Two vendors, one week apart, solved the same integration problem with opposite defaults. That is not a coincidence of engineering taste. It is a preview of how uneven MCP adoption is going to be across the entire martech stack, because nothing about the protocol itself requires read-only scoping, inherited permissions, or source citations. Those are choices a vendor makes, or skips.
This is not a two-company problem, either. Eventtia and Stravito are two data points inside a wave: research platforms, agency reporting tools and event and marketing operations vendors have all shipped MCP servers on their own schedules within the same stretch of weeks, each making its own call on scope and permissions with no shared baseline to check against. A marketing team connecting five or six of these tools to one AI assistant is not adopting a single governance posture, it is stacking five or six different ones, most undocumented, into a conversation with a model that cannot tell the difference between a read-only research citation and a live write to a payments record.
What it means for the marketing leader
This is exactly the gap this publication’s own reporting on GTM teams deploying AI agents faster than governance already surfaced: nearly a third of teams in that survey could not even quantify how many agents they had running, and sixty percent named ownership violations as a real fear, not a hypothetical one. An MCP server that grants an assistant write access to payments data is precisely the kind of deployment that survey warned about, arriving through a marketing tool’s plug-in page rather than through IT procurement. The same logic behind marketing automation trading static rules for AI judgment, as this publication covered when Phave launched its reasoning-based platform, applies here: judgment without a governance layer around it is not a feature, it is deferred risk.
The fix is not refusing every MCP connector until some industry-wide permission standard arrives, that standard does not exist yet and waiting for it is not realistic. The fix is asking one specific question before enabling any of these servers: does this connection inherit our existing permission model, or does it create a new one? Stravito’s answer to that question is in its own announcement. Eventtia’s is not, and a marketing operations leader who cannot get a straight answer to it from any vendor this quarter should treat that silence as the answer.
