MCP Servers for Marketing: Email, SEO, Analytics — What Is Available in 2026

Two years after Anthropic introduced MCP, the ecosystem is uneven. Deliverability, SEO and analytics are mostly covered. CDPs are patchy. ESPs are surprisingly thin. This article walks through every category that matters to a marketing engineer in 2026, and calls out what is actually usable versus what is a README on GitHub.

Scope of this inventory

Categories covered: deliverability, SEO, web analytics, CDPs, ESPs, CMS. Out of scope: ads platforms, attribution, CRM (sales-side). We will come back to those once the ecosystem catches up.

One-paragraph MCP refresher

MCP is Anthropic's open protocol that lets an LLM client (Claude Desktop, Cursor, Zed, custom agents) talk to tool servers over JSON-RPC. Each server exposes a handful of typed tools (start_test, get_report, etc). The LLM picks which tool to call based on the user's prompt. The result: natural-language workflows that span multiple products, without you writing integration code.

Category: email deliverability

The most mature MCP category for marketing. Testing email placement, checking authentication, reading DNS — all of these translate cleanly to tools, and the results are useful to a reasoning LLM.

Inbox Check MCP (ours)

Package: ldm-inbox-check-mcp on npm. Four tools: start_test, get_test, list_test, list_providers. Runs a real placement test against 20+ seed mailboxes (Gmail, Outlook, Yahoo, Mail.ru, Yandex, GMX, ProtonMail, iCloud, and more), returns SPF/DKIM/DMARC verdict, SpamAssassin score, per-provider folder placement. Free tier. See the automate-inbox-placement-tests-ai-mcp-claude guide for setup.

Others in this space

There are community wrappers for individual tools (e.g. Mailtrap, Mailhardener) but as of mid-2026 none of them expose the full authentication-plus-placement loop in one server. Most are "call this REST endpoint" thin wrappers without opinionated tool descriptions — which matters because a vague tool description produces vague agent behaviour.

Category: SEO

SEO is catching up fast. The dominant pattern is "expose the big three data APIs" (Google Search Console, Ahrefs, Semrush) plus a few crawl/audit servers.

Google Search Console MCP — community servers wrap the Search Console API. Lets the agent pull top queries, CTR, index coverage. Authentication is the painful bit — you need a service account.
Ahrefs MCP — hypothetical at the time of writing (a community server exists but hits rate limits fast on the paid API). If/when Ahrefs ships an official MCP, expect it to be one of the best in the ecosystem.
On-page auditors — Lighthouse MCP (by Google) is stable and genuinely useful. Point it at a URL, get a structured audit back, ask the LLM to prioritise fixes.
Crawlers — small community crawler MCPs exist but are slow for anything beyond a few hundred pages. For serious SEO work, still run a dedicated crawler (Screaming Frog, Sitebulb) and feed its export to the agent.

{
  "mcpServers": {
    "inbox-check":   { "command": "npx", "args": ["-y", "ldm-inbox-check-mcp"] },
    "gsc":           { "command": "gsc-mcp",    "env": { "GSC_KEY_FILE": "~/.config/gsc.json" } },
    "lighthouse":    { "command": "lighthouse-mcp" }
  }
}

Category: web analytics

GA4 is where most marketing data lives. The MCP story here is functional but limited.

GA4 Community MCP — wraps the GA4 Data API. Lets the agent ask "top landing pages last 30 days" or "conversion rate by source". Rate limits matter — GA4 has strict quotas per property per day.
Plausible MCP — an unofficial server exists for Plausible Analytics. Clean API, simple tool surface. Useful if you already use Plausible.
Matomo MCP — available. Less-used but works for on-prem Matomo deployments where you want to keep analytics data internal.

A note on data freshness

GA4 data lags. Intraday reports are best-effort. If you wire GA4 into an agent and ask "what happened in the last hour", the answer is usually "we do not know yet". Set expectations with the agent via the system prompt.

Category: customer data platforms

CDPs should be a natural fit — they are already APIs of events and profiles — but adoption has been slow.

Segment MCP — a community server wraps the Segment Public API. You can query workspaces, sources, destinations, recent events. Write operations are limited, which is probably correct for safety.
Rudderstack MCP — early-stage, community-maintained. Covers source/destination listing; not great for querying events.
PostHog MCP — surprisingly mature. PostHog ships an official MCP server that covers events, funnels, and feature flags. A good pattern for other vendors to follow.

Category: ESPs and marketing automation

This is where the ecosystem is thin. The vendors that should have built MCP servers first — Mailchimp, HubSpot, Klaviyo, ActiveCampaign — are mostly not there yet in 2026.

HubSpot — a community MCP exists, covers contacts and deals, but marketing email objects are partially supported. Expect it to improve.
Mailchimp — no official MCP. A couple of thin community wrappers around the REST API.
Klaviyo — community MCP covers lists and flows. Useful for segmentation questions, less for send-time performance.
Postmark / SendGrid — nothing official, but the API surfaces are small enough that a 200-line custom MCP server will get you 80% of the value.

Why the gap?

ESPs see the LLM-agent market as a distribution risk. If a user can ask Claude to switch providers in one sentence, lock-in disappears. Expect official MCPs from the incumbents only when they feel defensive, not proactive.

Category: CMS

Better coverage here than ESPs, because CMS APIs are simpler and the LLM use case is obvious (write draft posts, update metadata).

WordPress MCP — several community servers. Quality varies wildly. Pick one that uses application passwords, not a hard-coded admin token.
Webflow MCP — community server covers CMS collection items and published pages. Stable.
Contentful MCP — official SDK makes building this trivial; there is a good community MCP already in use.
Ghost MCP — works well for newsletter-focused blogs. Covers posts, members and newsletters.

Practical agent workflows combining multiple servers

Single-server agents are useful. Multi-server agents are where this gets interesting. A few concrete examples.

Workflow 1: pre-send deliverability sweep

Servers: Inbox Check + Contentful (or your CMS of choice).

Prompt: "Pull the draft newsletter from Contentful entry abc123, run a placement test from news.mybrand.com, and tell me if it's safe to send."

The agent fetches the draft HTML, runs start_test, polls get_test, reads the verdict, and either gives a green light or points at the broken auth record. 30 seconds end-to-end.

Workflow 2: post-send performance + deliverability

Servers: Inbox Check + GA4 + PostHog.

Prompt: "Compare last week's newsletter to this week's: inbox rate, click-through in GA4, and PostHog conversion funnel."

The agent fans out three tool calls in parallel, joins on the campaign id, gives you one paragraph of analysis.

Workflow 3: SEO-aware email topic picker

Servers: Google Search Console + your CMS + Inbox Check.

Prompt: "What are the top-trending organic search queries for our site this month that we have not covered in a newsletter yet?"

Query GSC, diff against newsletter archive in CMS, produce a ranked list, optionally draft a template and run a pre-send placement test.

What is still missing

Ads platforms. Google Ads, Meta Ads, LinkedIn Ads — essentially nothing official. Some community wrappers, none production-ready.
Attribution. No good MCP for Dreamdata, Hightouch Personalization, or similar. High-value target for whoever builds it.
Data warehouses as marketing primitives. There are Snowflake and BigQuery MCPs, but they are general-purpose. An opinionated "marketing warehouse" MCP that knows about the dbt marketing-analytics package would be useful.
Review and reputation platforms. G2, Capterra, Trustpilot — no MCPs at all as of mid-2026.

How to pick a server

Prefer official servers. They survive platform API changes.
Read the tool schemas before installing. Vague descriptions produce vague agents.
Check auth model. Servers that demand admin tokens are security debt.
Try it behind a read-only filter first. Some servers expose write tools by default that you do not want the agent to have.

Frequently asked questions

Is there a central registry of MCP servers?

Yes, there are multiple: the official MCP registry, community-maintained lists on GitHub, and several third-party hub sites. Quality varies. Treat any server you install with the same care as an npm dependency.

How do I run many servers in one Claude Desktop config without slowing it down?

Each MCP server is a subprocess. A handful is fine (say, up to 10). Beyond that, startup latency stacks up. Group related ones into a self-hosted gateway server if you want more.

Can the agent figure out which server to call on its own?

Usually yes, if the tool descriptions are good. That is the whole point of the protocol. If you find the agent picking the wrong tool, the fix is almost always better prompts or clearer tool descriptions, not more code.

Should I build my own MCP server for our internal marketing stack?

If your team uses an internal tool daily and it has a clean API, yes — a focused MCP server is a few hundred lines and pays back quickly. Start with read-only tools, add writes behind a confirmation pattern.

MCP servers for marketing: 2026 inventory