What is MCP in marketing?

MCP is the Model Context Protocol, an open standard introduced by Anthropic in November 2024 that lets an AI assistant connect to external tools and data through one described interface instead of a custom integration per tool. In marketing it means an assistant such as Claude Code can query analytics, read a CRM, or, with a work server like this one, take a launch plan and a directory index as tools and act on them.

What does a marketing MCP server do?

It exposes tools an AI client can call. A data server exposes reads: sessions from Google Analytics, spend from an ad account, contacts from a CRM, keywords from an SEO suite. A work server exposes the work itself: which step comes next, where a submission happens, what a form needs, and a way to record what was done so it can be checked.

Which MCP servers do marketers use?

Most stacks start with data: an analytics server, the biggest ad platform, the CRM, sometimes an SEO suite and a docs tool. Those tell the agent what happened. This server is the complement: it tells the agent what to do next and lets it record having done it. Use both; the plan here even says which numbers to go and read.

Is there an MCP server for Claude Code that does marketing?

Yes. Add https://marketinque.com/mcp with one command and Claude Code lists the 379-step launch plan and the 958-row directory index as tools, plus two prompts it exposes as slash commands. No signup for the free tier; a marketinque project adds the tools that record ticks, listings and ideas to a console.

Is the marketing MCP server free?

Yes, and it stays free. The free tier needs no account and has no usage cap you will meet by hand: the plan, the index, the submission pages, the kit draft and the two prompts. Progress lives in a marketinque.json in your own repository. A marketinque project — the console, the nightly checks, the idea queue — is free for now too, and it is opening by waitlist.

What does it cost to run?

Only what your Claude Code already costs. Every token of thinking — reading the plan, drafting the kit, filling a form, writing an idea — is spent in your Claude Code, on your Claude subscription or API key. This server stores state, runs the checks and holds the gates; it runs no model on your behalf, so marketinque never bills you for the drafting. A project can additionally hold a key of yours (Anthropic or Kimi) so the engine drafts between your sessions; that bill goes to your provider, and the only calls on marketinque’s own model are the two during setup. A project costs $0 for now, with no card asked for; joining the waitlist commits you to nothing.

Which Claude plan do I need?

Any plan that includes Claude Code: Pro, Max, Team or Enterprise, or an API key with Claude Code pointed at it. A backlink session of ten directories is a few hundred thousand tokens of reading, drafting and browser actions, which fits comfortably in a Pro session; a nightly routine on a project is what Max is for. Any other MCP client that speaks Streamable HTTP can use the same URL with whatever model it runs.

How do I get a marketinque project?

Join the waitlist — on this page, or by telling Claude Code you want one (it has a join_waitlist tool that asks for your email and your yes). Projects are opening in order; you get one email when yours is ready. Setup is two steps: your URL, then the connection — the project mints the token, gives you the one line to paste, and confirms the moment Claude Code first calls in.

Can Claude Code submit my product to directories by itself?

It can open the submission page, fill the form from your kit and read back the URL the listing lands at. It will not create accounts, type passwords, solve CAPTCHAs or, by default, click submit: it hands the browser back to you for each of those. Expect to press submit yourself, once per directory, with everything else already filled in and recorded.

Do I need a developer to use an MCP server?

Not for this one. Connecting is one command in Claude Code, or pasting the URL into any MCP client that speaks Streamable HTTP. There is nothing to install or host: the server runs at marketinque.com and answers over HTTPS.

How is this different from an MCP server for SEO tools?

An SEO suite’s server reports rankings, backlinks and keyword volumes about a site. This server is about earning the links in the first place: the ordered list of places that give one, graded by whether the link passes equity and whether it costs money, with the submission page on each and a record that checks itself. Read your numbers with the first; move them with the second.

mcp for marketing

MCP for marketing: give the agent the work, not only the data

Most marketing MCP servers connect an AI assistant to numbers: analytics, ad spend, a CRM. This one connects Claude Code to the work — a 379-step launch plan with prerequisites and 958 places that give a startup a backlink, as tools it can read, act on, and record against. Free, no signup, two lines to connect, and the thinking runs on the Claude subscription you already have.

By The Marketinque TeamReviewed September 2026

How it works The tools Free vs. project Questions

MCP for marketing, defined

MCP (Model Context Protocol) is an open standard, introduced by Anthropic in November 2024, that lets an AI assistant connect to tools and data through one described interface. MCP for marketing points that connection at marketing work: a server exposes what an agent can read (campaign data, rankings) or do (plan, submit, record), and any MCP client can use it.

The official MCP registry held 35,988 servers on 25 September 2026, counting each server once at its latest version. (Model Context Protocol registry, counted from its public API, September 2026.)

What you need

Three things, and the third is optional. Nothing to install on a server, nothing to host.

required
Claude Code
The terminal agent from Anthropic, on a Claude plan that includes it (Pro, Max, Team or Enterprise) or on an API key. That plan is what pays for the thinking — this server runs no model for you.
required
A product with a public URL
The agent reads your site to draft the kit and to fill every form. A landing page is enough; a waitlist page counts.
optional
A browser Claude Code can drive
For the form-filling: the Claude desktop app has one built in, and the Claude in Chrome extension lends it yours. Without one it still lists, drafts and records — you open the forms.

How it works, step by step

The server lives at https://marketinque.com/mcp, Streamable HTTP, stateless. Claude Code adds it with one command and lists its prompts as slash commands. The briefs behind the commands are served, not pasted, so they improve on deploy without you re-copying anything. Thirty minutes, first time through.

step 1
Add the server
One command, in the terminal, inside the repository of the product you are launching. Claude Code keeps connections per directory, so each product’s repo talks to its own.
claude mcp add --transport http marketinque https://marketinque.com/mcp
Or commit it to the repo: .mcp.json
```
{
  "mcpServers": {
    "marketinque": {
      "type": "http",
      "url": "https://marketinque.com/mcp"
    }
  }
}
```
Claude Code reads it on start, so everyone who clones the repo gets the server. With a project token (a project, below), keep the token in an environment variable the file refers to:
```
{
  "mcpServers": {
    "marketinque": {
      "type": "http",
      "url": "https://marketinque.com/mcp",
      "headers": {
        "Authorization": "Bearer ${MARKETINQUE_TOKEN}"
      }
    }
  }
}
```
Check: type /mcp in Claude Code and marketinque shows as connected, with the two slash commands under it. Any other MCP client that speaks Streamable HTTP takes the same URL.
step 2
Run the backlink brief with your product’s URL
This is the session most people come for: your product submitted to the free, dofollow directories, one form at a time, with you pressing submit. Add a number to cap the batch (… 10 does ten and stops).
/mcp__marketinque__get-backlinks https://your-product.com
What happens next, in the order it happens:
01 · claude code
Reads the kit, or drafts it
get_submission_kit returns the product profile and brand dossier for a connected project. Without one, the agent drafts the five things every form asks for from your site and shows them to you before using them.
02 · claude code
Lists the sprint
list_directories returns the 95 free, dofollow, page-scoped, graded rows in the order that works, each with its submission page where one is on record (90 are, fetched 2026-09-22).
03 · claude code
Opens the form and fills it
In the browser it drives, the agent opens the submission page, fills every field from the kit, picks the category, and attaches the logo where the form takes a URL.
04 · you
Sign in, pass the CAPTCHA, press submit
It hands the browser back for account creation, passwords, CAPTCHAs and the submit click, then continues when you say so. Most directories want at least one of those.
05 · claude code
Records where the listing landed
record_listing stores the URL and probes it: live and referencing your product counts as a referring domain. Free-tier sessions keep the same record in a marketinque.json in your repository.
Ten directories is a comfortable first session. The agent ends with a summary: submitted, skipped with the reason, and what needs you.
step 3
Work the launch plan
The same server holds the whole plan — 6 phases, 379 steps, each unlocked by the ones before it. This command asks for the next unlocked steps and packs a sprint into the minutes you give it (… 30 for half an hour). For each step it reads what, why, where it happens and how it verifies, does what it can from the terminal and the browser, and tells you what it cannot.
/mcp__marketinque__launch-plan https://your-product.com
Directories are one phase of the plan; the rest is positioning, the site, the first channel, measurement, the first customers. Run it whenever you have time to give it.
step 4
Keep the progress in your repository
On the free tier the server stores nothing about you. Every tick, every listing URL and every skip goes into marketinque.json in your repository, in a documented shape (the schema). Commit it. The next session hands the file’s done-list back to the server and starts exactly where this one stopped; the plan pages on this site can import the same file.
Inside a marketinque project the same tools record to the project instead, and the console and the nightly checks take it from there — the difference the next section is about. The project’s setup mints the token, hands you the one line to paste, and watches for the first call, so you know the connection took.

Two kinds of marketing MCP server

Every list of “best MCP servers for marketers” is a list of one kind. The other kind barely exists yet, and the two do different jobs.

data servers

What happened

Analytics, ad platforms, the CRM, an SEO suite, a docs tool. The agent reads sessions, spend, contacts and rankings and answers questions about them. Every stack should have two or three. They tell you where you are.

work servers

What to do next

A plan with prerequisites, so only unlocked steps are offered. Places to act, with the page where the action happens. A record of what was done that can be checked later. This is one of those, and it is free. It tells you where to go.

Use both. The plan here names the numbers to read (a domain rating of 20, thirty referring domains, a first channel that works) and the data servers are how the agent reads them.

What the agent gets

The same two pages this site publishes for people — the launch plan and the directory index — as tools an agent can call. Tool names as Claude Code shows them; the right-hand column says which tier has each.

get_planThe 6-phase launch plan as an outline, or every step under one phase or objective with its lock state.free
get_stepOne step in full: what, why, where it happens, the play it follows, what unlocks it, how it verifies.free
next_stepsThe next unlocked, unstarted steps in tree order, and a sprint packed into the minutes you have.free
list_directoriesThe 958 places that give a backlink, one phase at a time; the 95 free, dofollow, graded rows first.free
get_directoryOne directory: the numbers, the submission page, whether it wants an account first.free
get_submission_kitWhat every directory form asks for, written once: tagline, short and long descriptions, tags, story.free
set_traitsSay what kind of product it is — B2B, AI-native, serves an MCP, has a mobile app — so only the steps that apply are ever proposed.project
set_stepTick a step done, in flight, blocked or not applicable, on the founder’s say-so.project
snooze_stepPut a step off for “not yet”: by days, or until payments are live for the payment steps. It stays open in progress and comes back on its own.project
record_evidenceRecord the public URL a step’s proof lives at and probe it now; live and linking ticks the step.project
record_listingRecord what happened at a directory — listed with the URL, submitted and awaiting, or skipped with the reason — and probe it now; a live, linking page counts as a referring domain.project
verify_nowRun the sweep now: read the site, settle what it proves (traits, kit, listings, baseline), tick the steps that check themselves, re-probe every recorded URL.project
get_site_scanWhat the site scan saw — schema, socials, providers, pages, DNS, registry, day-0 mentions — so intake starts from facts.project
get_project_planProgress, the next steps, the sprint and the budget lens: the console’s plan page, as data.project
get_idea_briefThe exact prompt the engine gives its model for one idea type, built from your product, brand, facts and plan — so your own Claude Code drafts it instead, at no server cost.project
submit_ideaQueue what you drafted: validated (shape, placeholders, sources, novelty), then the same gateway and the same queue the engine fills.project
approve_ideaRecord the founder’s decision on a pending idea — approve, reject, snooze — through the console’s own approval path.project · approve scope
set_autonomyOn the founder’s word, change what runs without them: apply a preset, or give one action Ask me, a veto window, or Runs on its own. The same rules the console enforces; a gate is never raised.project · approve scope
get_autonomyEvery action that leaves the system, with its setting, what its rules allow and why not, and the current preset.project
claim_executionBefore posting an approved idea in the founder’s browser: log the claim, bound to a fingerprint of the approved text, and get a two-hour ticket; mark_posted with the URL and ticket then verifies the live page.project · approve scope
get_handoffAn approved idea as the founder would paste it: content and steps with the tracked link swapped in, the target, the contact.project
submit_brandSave a brand dossier you researched and wrote from the engine’s own brief. Every later draft grounds on it.project
report_signalReport a number from a data server you have attached (Search Console, Ahrefs, Stripe…) against a plan step; it ticks when the rule is met.project
submit_cycle_planPlan a strategist cycle from the same context the engine uses and submit it; the policy matrix filters it exactly as it filters the engine’s.project
session_summaryLeave what a session did, for the console and the founder’s morning email — what a nightly routine ends with.project

Forty-six tools in all, eleven prompts (get-backlinks, launch-plan, intake, ideas, review, post, brand-dossier, kit, verify, cycle, nightly) and two resources (the plan outline, the sprint). Plan version 50; the index is captured from seomade.app and graded there.

What is free, and what a project adds

The free tier is the server as a stranger meets it: no account, no token. A project is the same server connected to a marketinque account, where the ticks land somewhere that checks them.

free · no account

For any founder with Claude Code

The 379-step launch plan, with prerequisites, so only unlocked steps are offered
The 958-row directory index, the 95-row free sprint first, 90 submission pages on record
The get-backlinks and launch-plan briefs as slash commands
A drafted submission kit from your own site
Progress in a marketinque.json you own, in your repository
No account, no key, no limit on how many products

Free today, free when projects open. Start above.

project · opening by waitlist

For a product you are actually launching

A two-step setup: paste your URL and the overview of your product and your first X post are drafted for you; then connect Claude Code, an API key, or both — the console checks each connection live
A console that shows the plan with your ticks on it, and a nightly sweep that confirms each listing is still live and still links
The idea queue: researched, ready-to-post ideas — drafted by your Claude Code from the engine’s own briefs, or by the engine itself on a key you save (Anthropic or Kimi) — approved by you in the console
A saved submission kit and brand dossier every later draft grounds on
Reported signals: your data servers’ numbers (Search Console, Ahrefs, Stripe) tick the plan steps that ask for them
The nightly routine: one session a night that reviews, drafts, verifies and leaves a summary for your morning email
An audit log of everything the agent recorded, stamped as the agent’s, never silently

Everything in the free tier, plus the above, connected with a per-project token you mint in the console.

What it costs to run. The thinking happens in your Claude Code, on your plan. The server stores the state, runs the checks and holds the gates; it runs no model on your behalf, so marketinque never bills you for the drafting, on either tier. A project’s engine can also draft on the server between your sessions, on a key you save — Anthropic or Kimi, your key, your provider, your bill — but the same briefs let your Claude Code do that work on your subscription, which is the point of the design. The only two calls marketinque’s own model makes for a project are during setup: the overview of your site and your first X post.

What a project costs. Nothing, for now: a project is $0 and marketinque asks for no card (pricing). Joining commits you to nothing; you get one email when a project is ready for you, with the token that connects it.

How to join. The form at the bottom of this page, or tell Claude Code you want a project once it is connected: it asks for your email and your yes, and puts you on the same list.

what it will not do

A form filler with a conscience

No accounts, no passwords, no CAPTCHAs. Those come back to you, every time. A directory that needs a login says so on its row before the agent opens it.
No submit click by default. The agent fills; you send. The brief says this in its first paragraph, because a page that promised an autopilot and delivered a permission prompt would be the page misleading you, not the tool.
No paying, no fabricating. A paid placement, an invite-only community, a category you are not: skipped with the reason, and said in the summary.
No ticks it cannot back. A recorded listing is probed; a page that is live but hides its link stays “watching”, never “done”. Bought links that pass ranking credit are link spam under Google’s policy, and the index grades paid rows as such.
No decisions on its own. On a project, approving an idea, rejecting one or marking it posted needs both a scope you grant the token and your word in the session. The audit log says which it was.

Why a protocol, and not an integration

An integration is a promise between two products. A protocol is a promise to everyone: a server described once is usable by every client that speaks it, and a client that learns it can use every server. That is why this server is a URL and not a plugin — Claude Code today, whatever you run next year, the same tools with the same names.

It is also why the brief is served rather than pasted. When the plan changes (version 50 today) or a directory moves its form, the tools change with it, and nobody re-copies anything.

with Claude Code

Let Claude Code work this page

Connected, the index becomes a list it can work and the brief a slash command. It opens each submission page, fills it from your kit, and records the URL the listing lands at. It hands the browser back to you for accounts, CAPTCHAs and the submit click — and records where each thing landed.

1 · Connect, once

claude mcp add --transport http marketinque https://marketinque.com/mcp

2 · Then, in Claude Code

/mcp__marketinque__get-backlinks https://your-product.com

Free and read-only: progress lives in a marketinque.json in your repo. Inside a marketinque project the same tools record ticks on your plan and check each listing nightly; projects open by waitlist.

MCP for marketing, asked directly

What is MCP in marketing?: MCP is the Model Context Protocol, an open standard introduced by Anthropic in November 2024 that lets an AI assistant connect to external tools and data through one described interface instead of a custom integration per tool. In marketing it means an assistant such as Claude Code can query analytics, read a CRM, or, with a work server like this one, take a launch plan and a directory index as tools and act on them.
What does a marketing MCP server do?: It exposes tools an AI client can call. A data server exposes reads: sessions from Google Analytics, spend from an ad account, contacts from a CRM, keywords from an SEO suite. A work server exposes the work itself: which step comes next, where a submission happens, what a form needs, and a way to record what was done so it can be checked.
Which MCP servers do marketers use?: Most stacks start with data: an analytics server, the biggest ad platform, the CRM, sometimes an SEO suite and a docs tool. Those tell the agent what happened. This server is the complement: it tells the agent what to do next and lets it record having done it. Use both; the plan here even says which numbers to go and read.
Is there an MCP server for Claude Code that does marketing?: Yes. Add https://marketinque.com/mcp with one command and Claude Code lists the 379-step launch plan and the 958-row directory index as tools, plus two prompts it exposes as slash commands. No signup for the free tier; a marketinque project adds the tools that record ticks, listings and ideas to a console.
Is the marketing MCP server free?: Yes, and it stays free. The free tier needs no account and has no usage cap you will meet by hand: the plan, the index, the submission pages, the kit draft and the two prompts. Progress lives in a marketinque.json in your own repository. A marketinque project — the console, the nightly checks, the idea queue — is free for now too, and it is opening by waitlist.
What does it cost to run?: Only what your Claude Code already costs. Every token of thinking — reading the plan, drafting the kit, filling a form, writing an idea — is spent in your Claude Code, on your Claude subscription or API key. This server stores state, runs the checks and holds the gates; it runs no model on your behalf, so marketinque never bills you for the drafting. A project can additionally hold a key of yours (Anthropic or Kimi) so the engine drafts between your sessions; that bill goes to your provider, and the only calls on marketinque’s own model are the two during setup. A project costs $0 for now, with no card asked for; joining the waitlist commits you to nothing.
Which Claude plan do I need?: Any plan that includes Claude Code: Pro, Max, Team or Enterprise, or an API key with Claude Code pointed at it. A backlink session of ten directories is a few hundred thousand tokens of reading, drafting and browser actions, which fits comfortably in a Pro session; a nightly routine on a project is what Max is for. Any other MCP client that speaks Streamable HTTP can use the same URL with whatever model it runs.
How do I get a marketinque project?: Join the waitlist — on this page, or by telling Claude Code you want one (it has a join_waitlist tool that asks for your email and your yes). Projects are opening in order; you get one email when yours is ready. Setup is two steps: your URL, then the connection — the project mints the token, gives you the one line to paste, and confirms the moment Claude Code first calls in.
Can Claude Code submit my product to directories by itself?: It can open the submission page, fill the form from your kit and read back the URL the listing lands at. It will not create accounts, type passwords, solve CAPTCHAs or, by default, click submit: it hands the browser back to you for each of those. Expect to press submit yourself, once per directory, with everything else already filled in and recorded.
Do I need a developer to use an MCP server?: Not for this one. Connecting is one command in Claude Code, or pasting the URL into any MCP client that speaks Streamable HTTP. There is nothing to install or host: the server runs at marketinque.com and answers over HTTPS.
How is this different from an MCP server for SEO tools?: An SEO suite’s server reports rankings, backlinks and keyword volumes about a site. This server is about earning the links in the first place: the ordered list of places that give one, graded by whether the link passes equity and whether it costs money, with the submission page on each and a record that checks itself. Read your numbers with the first; move them with the second.

Want a project, when they open?

The free tier stays free — start with it today. A project adds the console, the nightly listing checks and the idea queue, on the same two commands. Leave your email and you hear once, when yours is ready.

Or start with the index, no account

Sources

Model Context Protocol: specification and documentation (modelcontextprotocol.io)
Anthropic, Introducing the Model Context Protocol (2024)
Anthropic, Claude Code documentation: connect Claude Code to tools via MCP
SEOmade: the startup directory index (seomade.app, 2026)
Google, Spam policies for Google web search (link spam) (Google Search Central)

Compiled by The Marketinque Team to our editorial standards.

MCP for marketing, defined

What you need

Claude Code

A product with a public URL

A browser Claude Code can drive