Does It Have an MCP Server? The One Question First Marketing Leaders Should Add to Every Martech Evaluation

You are building or rebuilding the marketing stack at a Series A company, and your team already lives in Claude or ChatGPT for most of the day. Every tool you evaluate this year will be used one of two ways: through its own interface, with the clipboard as the bridge to the assistant, or through a connector, with the assistant as the interface and the tool as the backend.

By the end of the first quarter, the tools in the first category are the ones your team copy-pastes out of. That is the whole argument for adding one question to every evaluation: does it have an MCP server, and what can the server do? This post gives you the seven-line version of that question, an honest scorecard for five tools your shortlist probably includes, and a fifteen-minute check you can run on anything else.

Why this moved to the top of the list

MCP, the Model Context Protocol, is the open standard that Claude, ChatGPT, Claude Code, Codex and Cursor all use to connect an assistant to another system. In two years it went from a developer curiosity to the plumbing under every “connector” and “plugin” the providers ship. The directory site mcpservers.org lists more than two hundred servers under marketing alone.

More telling than the count is who is talking about it. MKT1’s MCP showcase, a walkthrough of real GTM workflows built by marketers on connectors, includes a line from Zapier’s presenter that should be printed on every evaluation form: you mentioned pasting API keys into Claude, and you should not ever do that. That sentence contains two of the seven lines below.

Both providers are now pointed at your job. Anthropic’s Marketing plugin for Cowork wires Claude to HubSpot, Klaviyo, Amplitude, Ahrefs, Canva and Figma through MCP. OpenAI’s Codex shipped role-specific plugins in June 2026, with Sales (HubSpot, Salesforce, Clay) and Creative Production first and a Marketing Strategy plugin listed as coming, per Enterprise DNA’s coverage. The assistant is becoming the interface. The tools that plug into it well are the ones that will still be in your stack next year; the ones that do not join the list in why marketing teams end up over-tooled.

The seven lines

“Does it have an MCP server” is a yes-or-no question, and a yes is not enough. Here is the version that separates a checkbox from a tool you can actually run through an assistant.

1. Remote server over HTTP. Not a local process you install per laptop. A URL your whole team can connect to from any client.
2. Per-user OAuth sign-in. You connect by signing in with your own login in a browser, and the assistant sees what your account sees. No API key pasted into a chat, a config file or a plugin setting.
3. Reads that cover what marketing needs. Brand and voice, the current offer, the content calendar, campaigns and their pieces, automations and their plans. A server that only reads contacts is a CRM server, whatever the vendor calls it.
4. Writes that land as drafts. Every create or update should produce something a person reviews and publishes inside the tool. No send, launch, publish or delete tool in the connector. This is the line that matters most, and the reasoning is in the draft-only rule.
5. An audit trail. Every draft records who created it and through which assistant, so the queue is reviewable and the chat logs do not have to be.
6. Works across every client your team uses. Claude web and desktop, Claude Code, ChatGPT, Codex, Cursor. One server, many clients, no client-specific version.
7. Handles more than one account. For teams running two brands, or for an agency contractor, the connector should switch accounts without a second login.

Score a tool on these seven and you know, before the trial starts, whether your team will use it through the assistant or around it.

The scorecard

Five tools that show up on most Series A shortlists, scored on the seven lines from public documentation as of September 2026. Where the documentation does not say, the cell says so rather than guessing.

Two patterns in that table are worth naming. The incumbents lean read-only on marketing objects, which is safe and limiting. The challengers lean live, which is powerful and the thing line four exists to catch. Only one column is full-write and structurally draft-only, and we built it that way on purpose. Independent builders keep arriving at the same design; one developer who wrote his own HubSpot marketing email server left send out deliberately to prevent accidental sends. Klaviyo, ActiveCampaign, Customer.io and Brevo all have official servers, per Unspam’s comparison of email MCP servers; the differences between them are almost entirely in lines two through five.

What the plugins change, and what they do not

The Cowork Marketing plugin and the Codex role plugins are real improvements and worth installing. They are also generic until they can reach your system of record. The Cowork plugin keeps brand voice as a local setting rather than reading it from anywhere; its connectors are analytics, CRM and design tools; it produces plans and drafts as documents. The longer review of what it still cannot see goes through the week-two problem in detail. Codex’s Sales plugin reaches the CRM; a Marketing plugin is announced, not shipped.

The pattern is that the providers ship the assistant and the skills, and the marketing platforms are expected to ship the server. So the evaluation question is not “does Claude or ChatGPT support this tool.” It is “does this tool support the assistant well,” which is the seven lines. Cursor’s marketplace already carries a HubSpot plugin, and Composio is publishing app guides for marketers on Cursor; the clients are multiplying, which is why line six is on the list.

The fifteen-minute check

For any tool not in the table, this is enough to score it before a sales call.

1. Search the vendor’s docs for “MCP”. If the only hit is a community project or a transactional sub-product, line one is a no.
2. Read the tool list. Every MCP server publishes its tools’ names and descriptions. Sort them into reads and writes, and look for anything that can send, launch, publish or delete. That is line four, answered in two minutes.
3. Look at the auth screen. Connect it once. If the setup asks you to paste a key, line two is a no. If a browser sign-in opens with your own login, it is a yes.
4. Run one prompt. “Tell me what is on the calendar this month and what automations are live.” If the answer is contacts and nothing else, line three is a no.
5. Ask the vendor for lines five and seven. Audit trail and multi-account are rarely on a docs page and always on a security questionnaire.

The check is the same one your security reviewer will run later, which is why it belongs in the evaluation now. The stack that survives consolidation, described in martech stack consolidation at Series A, is the one where every tool passes it; the stack that gets rebuilt is the one where the assistant became the interface and half the tools could not follow.

The worked example

Marqeable is the column we can vouch for, so here is exactly how it scores and how to check it yourself. The server is https://app.marqeable.com/mcp, a remote MCP server over HTTP. Sign-in is OAuth 2.1 through your normal Marqeable login in a browser window, so the assistant sees what your account sees and no key is stored anywhere.

Reads: ten tools covering who you are, your business context (brand, ICP, voice, offers), campaigns with their pieces and send days, semantic search over existing content, the content calendar with scheduled sends and market dates, and every automation with its plan. Writes: six tools that create campaign drafts, content drafts, anchored suggestions on existing copy, brief updates and automation drafts. There is no send, launch, publish or delete tool. Every draft records who created it and through which assistant, and every result links to the page in Marqeable where a person approves it. Teams with several accounts list and switch between them from the assistant.

To connect, as of September 2026: in Claude, Settings > Connectors > Add custom connector with the address, then Connect and sign in. In Claude Code, claude mcp add --transport http marqeable https://app.marqeable.com/mcp, then /mcp to sign in. In ChatGPT, Settings > Connectors > Advanced > Developer mode, then create a connector with the address and OAuth. In Codex, add the server to ~/.codex/config.toml and run codex mcp login marqeable. In Cursor, add it to .cursor/mcp.json. The first prompt on any of them: “Run whoami, then get_business_context, then tell me what is on the calendar this month.” Marqeable syncs audiences from HubSpot, Salesforce and ServiceTitan and complements the CRM rather than replacing it, so the HubSpot column and this one belong in the same chat; that setup is walked through in connect HubSpot to ChatGPT. If you are also thinking about whether your own product gets recommended by the assistants, that is a separate question, covered in how to get your SaaS recommended by ChatGPT.

Frequently asked questions

What is an MCP server and why does it matter for a marketing tool?

MCP, the Model Context Protocol, is the open standard that Claude, ChatGPT, Claude Code, Codex and Cursor use to connect an assistant to another system. A marketing tool with an MCP server can be read from and, if it allows writes, drafted into from the assistant your team already uses. A tool without one can only be used through its own interface and the clipboard.

Is having an MCP server enough?

No. The useful questions are whether it is a remote server that signs you in per user with OAuth rather than a pasted API key, whether its reads cover the things marketing needs such as brand, calendar and campaigns, whether its writes land as drafts or go straight to live, whether it keeps an audit trail, whether it works across every client your team uses, and whether it handles more than one account.

Which marketing tools have MCP servers as of September 2026?

HubSpot has a remote MCP server and connectors for Claude and ChatGPT. Klaviyo, ActiveCampaign, Customer.io and Brevo have official servers. Mailchimp’s only official server covers its transactional email product, not campaigns. Marqeable has a remote server with draft-only writes. Details change monthly, so check each vendor’s docs.

Should draft-only writes be a hard requirement?

For any tool that can reach a customer, yes. A connector with a send or launch tool turns an assistant mistake, a stale instruction or a poisoned web page into a message to your list. A connector whose every write is a draft turns the same mistake into a queue item. This is also the only design that is safe in unattended runs such as scheduled routines.

The bottom line

The assistant is the interface now, and the platform is the backend. A marketing tool without a good MCP server is a tool your team will be copy-pasting out of by the end of the quarter, whatever its feature list says. Add the seven lines to every evaluation, run the fifteen-minute check before the sales call, and weight line four above the rest: reads should be broad, writes should be drafts, and nothing in the connector should be able to reach a customer without a person in the loop.

See a full-score connector: Marqeable’s agents build and run your campaigns, run behavior-triggered automations across text and email, answer and qualify every visitor with AI website chat, and tie revenue to the exact message with attribution.

Marqeable runs your campaigns, answers every visitor, text, and email in seconds, and turns them into booked jobs and meetings - even at 9pm on a Saturday. We’re in private beta with a small early cohort. Get early access