Claude Code for Marketing: Skills, a Context File, and One MCP Server

If you are a technical founder doing your own marketing, or the first marketing hire at a company where the founder already lives in Claude Code, you have probably discovered that the terminal is a surprisingly good place to write a launch email. Marketing skills give it a playbook. A context file gives it your product. The drafts come out fast and mostly right.

And then every one of them ends up as a markdown file that you carry, by hand, into the email tool, the social scheduler and the CRM. The playbook does not know what is on your calendar. The context file does not know which automation is live. The output has nowhere to go.

This post is the setup that closes that gap: skills for the how, a context file for the what, and one MCP server for the where, committed to the repo so everyone on the team gets the same connection. It takes one command, and everything the assistant writes lands as a draft you approve.

Skills are recipes

A Claude Code skill is a reusable instruction file that tells the assistant how to do a specific job. The best-known collection for this audience is Corey Haines’ marketingskills repository, which has collected tens of thousands of GitHub stars and covers the working marketer’s list: positioning, launch emails, landing page copy, pricing pages, cold outreach and more. Anthropic’s own team is a useful proof point; its write-up of how its marketing team uses Claude describes a growth marketer with no coding background building ad tooling in Claude Code.

A skill is exactly as good as a recipe. It captures a method that works and makes it repeatable. What it cannot capture is your ingredients or your kitchen: it does not know your product, and it does not know where the finished dish is supposed to go.

A context file is a pantry

The second layer solves the ingredients problem. The marketingskills repo, along with most guides to setting up Claude Code for marketing, converges on a “define once” pattern: a shared product-marketing context file in the repo, alongside CLAUDE.md, that holds your positioning, ICP, voice rules, current offer and the words you do not use. Every skill reads it. Nobody re-explains the company at the top of a session.

This is a real improvement and you should do it. It is also a document, with a document’s two limits. It goes stale the day the offer changes unless someone edits it. And it can only hold guidance. It cannot hold state: what is scheduled, what is running, what is waiting for review.

Neither is a kitchen

Run a launch-email skill against a good context file and you get five well-written emails in a markdown file. Now what?

The file does not know that a webinar reminder is already scheduled for the 16th. It does not know that a follow-up automation already sends a check-in on day 7, so email three duplicates it. It has no send day, no campaign it belongs to and no approval status. Somebody opens the email tool and starts pasting. That carry is where founder-led marketing stalls; the pattern shows up in why marketing ships slower than engineering and it is what vibe marketing has to solve before it produces pipeline.

The people building with Claude Code have arrived at the same conclusion. MKT1’s MCP showcase, a walkthrough of GTM workflows built by real marketers, sums it up: Claude becomes far more useful when it is connected to the other tools in your GTM stack. Vendors are following; Klaviyo published its own guide to using Claude Code with the Klaviyo MCP server in August 2026, aimed at marketers with no engineers. The missing layer is a system of record the skills can write into, with a calendar, automations and an approval step.

One command

Marqeable’s MCP server lives at https://app.marqeable.com/mcp. It is a remote server over HTTP with OAuth 2.1, so you sign in with your normal Marqeable login in a browser window and Claude Code sees exactly what your account can see. No API keys go in a config file or a chat.

Per Anthropic’s MCP documentation for Claude Code, as of September 2026 the setup is:

claude mcp add --transport http --scope user marqeable https://app.marqeable.com/mcp

Then, inside a Claude Code session, run /mcp, pick marqeable, and complete the sign-in in the browser. The --scope user flag makes the connection available in every project on your machine. Leave it off and the server is scoped to the current project.

To give the whole team the same connection, commit a .mcp.json at the repo root:

{
  "mcpServers": {
    "marqeable": {
      "type": "http",
      "url": "https://app.marqeable.com/mcp"
    }
  }
}

Anyone who opens the repo in Claude Code gets prompted to approve the server and signs in with their own login. The marketing repo now carries three things: the skills, the context file and the connection, and a new hire is productive on day one.

What is on the other end of the line: ten read 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. Six write tools create campaign drafts, content drafts, anchored suggestions on existing copy, brief updates and automation drafts. Every write is a draft. There is no send, launch, publish or delete tool, and every result carries a link to the page in Marqeable where a person approves it.

A session: skill in, automation draft out

Here is what the three layers look like working together. The skill supplies the method, the context file the voice, and the connector the state and the landing.

You: Run whoami, then get_business_context, and compare it with .agents/product-marketing.md. Tell me what is out of date.

Claude Code: Signed in as the founder on the Northwind account. Two differences: the account’s current offer is the 30-day pilot, while the context file still says free audit; and the account’s ICP names first controllers at 50 to 200 person companies, while the file says finance leaders. I can update the file to match the account.

You: Update it. Then use the email-sequence skill to write a five-step launch sequence for the pilot, and check the calendar and automations before you pick days.

Claude Code: Context file updated. The calendar has a nurture email on the 4th, the webinar reminder on the 16th and a LinkedIn post on the 18th. The pilot-offer automation is live with a day-7 check-in, so I dropped the check-in email from the sequence to avoid a duplicate. Sequence drafted: five emails on days 0, 3, 9, 13 and 20 starting the 8th, none on the 16th or 18th.

You: Make that an automation draft, quiet hours on, approval on the first email.

Claude Code: Created an automation draft with five steps at those intervals, inside the 7am to 9pm window, with an approval gate on step one. It is not live. Review it here: [link to the automation draft in Marqeable].

You: Last thing: the subject line on last week’s webinar invite was weak. Suggest a better one.

Claude Code: I found the invite in your content and added a suggestion anchored to the current subject line with a replacement and the reason. It shows up as a comment on the draft for you to accept or reject.

Four things happened that a skill and a file cannot do alone: the context file was corrected from the live account instead of from memory, the sequence avoided a real collision and a real duplicate, the output landed as an automation draft with timing and a gate, and the edit to existing copy landed as an anchored suggestion rather than a rewrite in a chat.

Where the assistant is allowed to write

Giving a terminal agent write access to the system that emails customers deserves a moment’s thought. Claude Code asks for permission on tool calls, and you will be tempted to allow them permanently after the tenth prompt. The safer layer is the server’s design: Marqeable’s connector cannot send, launch, publish or delete, so the worst outcome of any call is a draft in the review queue. That is what makes it reasonable to allow the connector’s tools without a prompt on every call, and it is the standard to hold every marketing MCP to; the longer argument is in the draft-only rule.

The same server works from Codex, Cursor and the Claude desktop app, which matters for founders who rotate tools. The connection is to the account, not to the client. If you want Claude Code to prepare the week’s drafts on a schedule rather than on demand, the next step is Claude Code Routines for marketing.

This is not an AI CMO

Corey Haines, whose skills you may be running, has been dismissive of most products sold as an AI CMO, and he is right to be. Strategy, positioning and the judgment about what to say to whom are not what this setup automates. What it automates is the distance between a decision and a draft in the right place: the carrying, the checking against the calendar, the rebuilding of timing in a workflow tool. That is the work that eats a founder’s evenings, and it is the part a founder acting as CMO most needs off their plate before the first marketing hire. Whether the next step is that hire or a GTM engineer, they inherit a repo where the skills, the context and the connection already work.

Frequently asked questions

Can I use Claude Code for marketing if I am not a developer?

Yes. Claude Code is a terminal assistant, but marketing skills and a context file turn it into a writing and planning tool that needs no code. Anthropic’s own marketing team uses it this way. The one technical step is a single command to connect it to your marketing system, after which it works in plain English.

What is the difference between a Claude Code skill and an MCP server?

A skill is a reusable instruction file that tells Claude Code how to do a job, such as write a launch email sequence. An MCP server is a live connection to a system, such as your campaign calendar and automations, that Claude Code can read from and write drafts into. Skills are the recipe; the MCP server is the kitchen the recipe is cooked in.

How do I connect Claude Code to Marqeable?

Run claude mcp add --transport http marqeable https://app.marqeable.com/mcp in your terminal, then run /mcp inside Claude Code, pick marqeable and sign in with your normal Marqeable login. To share the connection with your whole team, commit a .mcp.json file in the repo with the same URL.

Can Claude Code send a campaign from the terminal?

Not through Marqeable’s connector. It can read your business context, campaigns, content, calendar and automations, and it can create drafts of campaigns, content, suggestions, briefs and automations. There is no send, launch, publish or delete tool. Every draft waits for a person to approve it in Marqeable.

The bottom line

Skills are recipes and a context file is a pantry, and together they make Claude Code a good marketing writer. Neither is a kitchen. Until the assistant can read the calendar and the automations and hand its work back as a draft in the system that runs it, every skill ends in a file you carry by hand. One command and one committed config file fix that for the whole team, and because every write is a draft, the founder keeps the approval.

See the kitchen: 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