The Best MCPs for Marketing: Hyper vs GoMarble (2026)

Picking a marketing MCP in 2026 is harder than it should be. Dozens exist now. Some are open-source single-platform servers (GoMarble for Meta and Google, Pipeboard for Meta, Adspirer for paid). Some are first-party MCPs from the ad platforms themselves: Meta launched its server in April, Google has shipped one for read-only Google Ads access since October 2025, Amazon shipped in February. And one is a full marketing platform exposed over MCP, with the agent system, scrapers, and integrations baked in: Hyper.

Most comparisons skip the question that actually matters. What does the MCP let an AI agent do, and how much setup before that becomes clear? Reading data is one job. Launching a TikTok campaign with a 5,000 dollar daily budget is another. Posting organic content to Instagram is a third.

The rest of this post is a working comparison of Hyper and GoMarble, the two products that come up most when marketing teams ask which MCP to install in Claude or Cursor. The broader landscape shows up where it matters.

The full comparison

What it looks like in practice

A real prompt makes the difference clearer than a feature list does.

Marketer opens Claude, types: "Launch a TikTok campaign for our new winter product, 200 dollars a day, US iOS users 18 to 35. Then check Meta from last week and tell me which ad sets to scale."

With Hyper installed, Claude does both halves. It launches the TikTok campaign through the Hyper MCP, picks the campaign type, sets bids, configures targeting. Then it pulls the trailing 7 days of Meta performance and pulls out the ad sets paying for themselves against the CPA target. The next message is usually some version of "want me to scale these and pause those, or wait for approval?"

With GoMarble installed, Claude answers the second half. It can't launch the TikTok campaign. GoMarble's MCPs are read-only, and they don't cover TikTok in the first place.

What GoMarble's MCPs actually do

GoMarble runs three open-source MCP servers on GitHub: Facebook Ads, Google Ads, and Google Analytics 4. All three are read-focused. All three are free and MIT-licensed. The Facebook one is the most popular, sitting at 318 stars at time of writing.

The Facebook Ads MCP exposes 26 tools, all of them either get_X or list_X. You can ask Claude what your campaigns did last week, which ad sets crushed it, what hooks the top performers used. You can't ask Claude to launch a new campaign or pause an old one. That's outside the scope.

Same shape for the Google Ads and GA4 servers. Read-only, single-platform, narrow on purpose.

For developers querying ad data from inside Claude or Cursor, GoMarble's MCPs are a clean option. The price is zero and the code is auditable. The ceiling is read-only access. For most marketing teams running ads day-to-day, that ceiling shows up fast.

What Hyper's MCP exposes

Hyper exposes its full marketing platform over MCP. Install command is npx skills add hyperfx-ai/marketing-skills, then OAuth or paste an API key. About a minute. The same install works in Claude, Cursor, Codex, ChatGPT, OpenCode, GitHub Copilot, and Gemini CLI.

What's exposed comes in three buckets.

100 plus skills

Skills are reusable workflows that operate like an expert in a domain. The marketing skills include Voice DNA, CMO, Ad creative, SEO and GEO, viral content, Meta ads SOP, client reporting, and competitive research, among others. When the agent runs a skill, it uses the skill's playbook, not just the raw API.

The 17 skills shipped through hyperfx-ai/marketing-skills cover the core marketing surface: paid ads on Meta, Google, TikTok, LinkedIn, Amazon, Pinterest, organic social on Instagram, TikTok, LinkedIn, plus research, creative, lifecycle email, and analytics. The full Hyper platform has 100 plus skills across more domains.

40 plus built-in tools

Built-in tools come pre-installed. No extra keys, no packages, no separate setup.

• Social media scrapers: Instagram, TikTok, LinkedIn, Reddit, Twitter, YouTube, Facebook
• Meta Ads Library scraper for pulling competitor ads
• Web scraping (Firecrawl, Outscraper) and a web browser
• Web search (Tavily, Exa) and website analyzer
• HyperSEO for keyword research, AI search visibility, competitor analysis, rank tracking
• Image generation (GPT Image, Seedance)
• Video generation (Seedance 2.0, Google Veo, OpenAI Sora, HeyGen)
• Code sandbox for custom logic
• Database, vector store, knowledge base for the agent's memory
• YouTube transcripts, Google Trends, Google Search scraper, Crunchbase, SEC data

200 plus integrations

Integrations are how the agent connects to outside platforms. One click each. The current list spans paid ads, organic social, email and lifecycle, CRM, analytics, ecommerce, productivity, communication, and content management.

When the agent runs a paid-ads skill, it can pull GA4 data, Shopify orders, Klaviyo email performance, and Slack notifications from the same conversation.

The framing Hyper uses publicly: MCP servers give you the plumbing; Hyper gives you the house.

Run Meta Ads from Claude

Connect Hyper MCP to Claude with OAuth. Any prompt about Meta ads is now something Claude can act on directly.

Examples that work end-to-end:

• "Launch a Meta campaign for our new product, 500 dollars a day, target lookalikes of our top 10 percent customers."
• "What were my best three ad sets last week by ROAS? Pause anything below 1.5x."
• "Pull my top 50 competitors' Meta ads from the last 30 days and tell me which hooks they're using." (Uses the built-in Meta Ads Library scraper.)
• "Generate three new ad concepts in our brand voice based on what's working." (Uses the brand profile from the agent's research phase plus the creative generation tools.)
• "Set up an alert: ping my Slack if any campaign's CPA climbs past 35 dollars."

Ads Manager isn't required after that. The agent runs the actions through Hyper's connection to the Meta Marketing API.

Run TikTok Ads from Claude

Same shape, different platform. Hyper covers TikTok Ads natively, including Smart+ campaign types.

• "Launch a TikTok campaign for our spring drop, 200 dollars a day, US iOS users."
• "Pause any TikTok ad that hasn't dropped CPA below 25 dollars in 48 hours."
• "Generate five UGC-style ad briefs based on what's hitting on TikTok this week."
• "Pull what our three competitors are running on TikTok right now."
• "Post the new product launch organic to TikTok, write the caption in our voice, schedule for tomorrow at 9am."

Note the last one: organic posting from Claude. Hyper covers organic social as well as paid, which is uncommon in the marketing-MCP space. Most MCPs that exist today are paid-only or analytics-only.

Use it from ChatGPT, Cursor, or any AI client

Hyper's MCP works with every AI client that speaks MCP, plus the broader Agent Skills protocol. Claude (Desktop or API), Cursor, Codex, OpenCode, GitHub Copilot, Gemini CLI, and ChatGPT through Connectors.

Same skills, tools, and integrations show up across all of them. A founder might use Claude Desktop. A media buyer might prefer Cursor for its terminal-style flow. A developer building a marketing tool on top of Hyper might use Codex. An ops team might run Hyper directly through its own dashboard. The agent behaves the same way wherever it's running, because the underlying MCP is identical.

Setup in 60 seconds

For Hyper:

1. Run npx skills add hyperfx-ai/marketing-skills, or click the install link on hyperfx.ai/mcp
2. Connect ad accounts via OAuth, or paste an API key
3. Set goals (cost-per-lead targets, ROAS, monthly budget, brand-voice notes)
4. Open the AI client. Skills, tools, and integrations are ready.

For GoMarble:

1. Install Python 3.10 or higher
2. Set up a virtual environment
3. Get a Meta access token from the Meta Developer portal
4. Edit Claude Desktop config to point at the server
5. Repeat per ad platform (Facebook Ads, Google Ads, GA4 are separate MCPs)

Both work. The setup time is meaningfully different.

Who picks what

By the situation the team is actually in.

Built on serious infrastructure

Hyper runs on the same backbone enterprise SaaS runs on. Supabase handles the database and auth. Google Cloud Platform and AWS handle hosting. e2b runs the agent execution sandboxes. Hyper is part of the Nvidia Inception program for AI-native companies.

The data Hyper holds doesn't sit on a stack built from scratch. It sits on the same security and compliance backbone the enterprise tools you already use are built on. Connections happen through OAuth via business accounts, with token rotation and refresh handled by the platform. Sensitive actions can require human approval before execution.

A few things Hyper isn't

Worth being clear about scope, since "agents that run end-to-end" can sound like a black box.

Brand voice, strategic positioning, big creative bets all stay with the team. The agent works inside whatever framing it's given. If the team hasn't decided who the brand is, the agent can't either.

Regulated creative for alcohol, finance, or health claims routes to human approval before launch. No override.

Replacing an agency is a different conversation. Agencies do strategy, relationships, and the big creative work. Hyper takes the daily ops a junior media buyer or social manager would handle. Complementary scope, not a replacement.

The verdict

For developers querying ad data from Claude or Cursor: GoMarble works. They're free, focused, and well-maintained.

For brands, agencies, founders, and anyone running marketing day-to-day from an AI client: Hyper covers the wider surface, with skills, scrapers, integrations, and write actions all available from the same connection.

Start a Hyper trial at hyperfx.ai or run npx skills add hyperfx-ai/marketing-skills to install it.