The connectors worth a marketer's time are the ones that reach live account data and survive a real question. On September 14, 2026 we called every endpoint in this post directly; the twelve below answered, held an OAuth grant, and do work a marketer actually has. The rest of the directory is excellent software aimed at somebody else's job.
Last updated September 14, 2026. New to any of this? What MCP is explains the protocol underneath in plain English. This post is the shortlist.
What a Claude connector actually is
A connector is Anthropic's product word for a remote MCP server that Claude can reach over HTTPS. You authorise it once with OAuth; from then on Claude can see the list of actions that server offers and call them mid-conversation, acting with the permissions you already have in that tool.
Two things follow from that and neither is obvious:
- The directory is not the boundary. Anthropic curates a connectors directory inside Claude, but any remote MCP server can be added by URL as a custom connector whether or not it is listed (Anthropic). The directory is a trust signal, not a gate.
- The plan decides how many. Custom connectors exist on Free, Pro, Max, Team and Enterprise, and Free is capped at one. On Team and Enterprise an Owner adds the connector for the organisation and everyone else clicks Connect (Anthropic).
How we checked
Every endpoint below was sent a tools/list request over streamable HTTP on September 14, 2026. A live, correctly built remote server answers one of two ways: with its tool list, or with 401 and a WWW-Authenticate header pointing at its OAuth metadata. Both mean the server is up and the setup in its documentation is real. Anything that timed out or returned a generic 404 was cut, however well reviewed it is elsewhere.
One result is worth stating up front, because no other article on this SERP mentions it. Shopify's Storefront MCP documentation describes six tools. We called six live .myshopify.com endpoints; five answered without authentication, and every one of them returned exactly one tool — search_shop_policies_and_faqs. The catalogue tools are not on by default. Documentation is not a probe.
The twelve worth connecting
Ordered by how much of a marketing week they remove, not by popularity.
1. Sprites: the paid media and SEO layer
Full disclosure: we build this one. It is first because it is the only connector on this list that acts on ad accounts — Google Ads, Meta, LinkedIn, TikTok and Reddit, plus GA4, Search Console, Shopify and Google Business Profile, behind one endpoint. Ask it where Google Ads is wasting money and it reads the live account, does the analysis and hands back changes; every write arrives as an approval card you edit before it applies. The Claude setup is a URL and a sign-in.
Best for: anyone who runs paid media. Watch for: it acts on channels you have connected inside Sprites — it is not a generic API wrapper.
2. Semrush: keyword, competitor and traffic data
Semrush's official remote server is the single biggest upgrade to research-in-chat. It exposes the Standard API (keyword research, backlinks, organic and paid research, domain analytics), the Trends API for traffic and market estimates, and a read-only view of your Projects. The read-only Projects scope is the detail worth noticing: Semrush deliberately did not let a model reconfigure your tracking.
Best for: SEO and competitive research without leaving the thread. Watch for: it bills against your Semrush API limits, and a chatty model burns them fast.
3. HubSpot: the CRM Claude can actually read
HubSpot's server answers at a Claude-specific path and covers contacts, companies, deals and engagements. The marketing use is less "look up a contact" than "which of last quarter's closed-won deals came from the campaigns I am about to renew" — a question that normally means an export.
Best for: attribution questions that span CRM and campaign. Watch for: scope sprawl; grant it the objects you need and no more.
4. Notion: where the brief already lives
Notion's MCP server is the most used connector in most teams for a boring reason: the brief, the calendar and the messaging doc are already there. Claude reading them before it writes anything is the difference between on-brand copy and generic copy.
Best for: giving every other connector the context it lacks. Watch for: a large workspace is a lot of surface area — connect it, then be specific about which database you mean.
5. Canva: creative production, not creative judgement
Canva's server lets Claude find, generate and resize assets in your brand's own Canva account. It is genuinely useful for the mechanical half of creative work — the eleven sizes of one concept — and does not replace the decision about which concept.
Best for: resizing and versioning at volume. Watch for: brand kit discipline; an unconstrained generation is a design review you did not schedule.
6. Figma: for teams whose source of truth is a file
Figma's MCP server reads design context out of a file — components, variables, screenshots — which matters for marketers only when the landing page and the ad both come from the same design system. When they do, it is the connector that stops copy from being written against an out-of-date mock.
Best for: landing-page work alongside a design team. Watch for: it is aimed at the design-to-code path; the marketing value is a side effect.
7. Klaviyo: lifecycle and the revenue it actually drove
Klaviyo's server covers campaigns, flows, lists, segments and metrics. For ecommerce it closes a loop nothing else on this list closes: the same conversation can ask what paid spend did and what the flow behind it did.
Best for: DTC teams running email and SMS beside paid. Watch for: write scopes; a flow is not a thing to edit conversationally without a review step.
8. Shopify Storefront: live, free, and thinner than advertised
Every eligible store answers at {shop}.myshopify.com/api/mcp with no authentication at all — including your competitors'. That is the interesting part: you can ask Claude to read a rival store's published policies and FAQs in seconds. As noted above, in our probe each live store exposed only search_shop_policies_and_faqs, so treat the catalogue tools as opt-in rather than assumed. For your own store's performance data rather than its shop front, Shopify through Sprites is the authenticated route.
Best for: fast competitive checks. Watch for: it is the shopper-facing surface — no orders, no sessions, no conversion data.
9. Stripe: the number nobody argues with
Stripe's server gives Claude revenue, subscriptions, customers and refunds. Connect it beside an ads connector and "our ROAS" stops being a platform-reported figure and starts being a comparison against money that actually arrived.
Best for: reconciling reported ROAS against real revenue. Watch for: it can create objects. Read-only scopes unless you have a reason.
10. Webflow: publishing without a handoff
Webflow's server reaches sites, collections and CMS items, which makes it the shortest path from "this page needs a new meta description" to the page having one. The equivalent for WordPress is the Abilities API and the mcp-adapter plugin rather than a hosted connector — covered in the wider MCP server roundup.
Best for: small on-page changes at the speed you notice them. Watch for: publishing is a write. Stage it.
11. Zapier: the escape hatch
Zapier MCP advertises 9,000+ apps and 30,000+ actions, and you choose which of them a given server exposes. It is how you reach the tool nobody has built a connector for. The cost model is the catch and it is easy to miss: each tool call spends two tasks from the same quota your zaps use, so an exploratory conversation is not free. We took this apart in Zapier MCP, examined.
Best for: the long tail. Watch for: task burn, and tool sprawl — expose five actions, not fifty.
12. Slack: where the answer needs to land
Reading a channel for context and posting a summary back into it is unglamorous and it is what makes the rest of this list visible to anyone but you. Worth adding last, once something is producing answers worth sharing.
Best for: distribution of everything above. Watch for: posting scopes in a busy workspace.
Also live, if your stack needs them
Called on September 14, 2026; each answered with an OAuth challenge, which is what a healthy remote server does.
| Connector | What it reaches | Marketing use |
|---|---|---|
| Airtable | Bases, tables, records | Content calendars and campaign trackers already kept there |
| Linear | Issues, projects, cycles | Turning a campaign retro into tickets someone owns |
| Asana / Monday | Tasks, projects | The same, for teams that live in one of those instead |
| Atlassian | Jira, Confluence | Enterprise marketing ops and anything shared with engineering |
| PayPal / Square | Transactions, orders | Revenue truth where Stripe is not the processor |
| Intercom | Conversations, contacts | Mining support threads for objections worth answering in ad copy |
| Vercel / Cloudflare | Deployments, docs, analytics | Technical SEO and page-speed work on your own property |
The gap nobody fills
There is no official Google Ads connector, no official Meta Ads connector and no official Search Console connector. Google publishes an MCP server for Analytics — experimental, read-only, local-only — and stops. Meta publishes nothing.
So the paid-media category has exactly two options: a community wrapper you install, host and keep current against an API that changes quarterly, or a hosted product that holds the partner integration for you. That is the choice, and it is worth making deliberately rather than discovering three months in. The Google Ads MCP comparison and Meta Ads MCP comparison lay out both sides.
How to add a connector to Claude
1. Open Connectors in settings
Settings → Customize → Connectors on Free, Pro and Max. On Team and Enterprise, an Owner does this once under Organization settings → Connectors; everyone else then sees a Connect button (Anthropic).
2. Add the URL
Click +, then Add custom connector, and paste the server's HTTPS endpoint. A well-built remote server publishes its own OAuth client metadata, so there is nothing to fill in under Advanced settings. If a server asks you for a client ID and secret, that is a 2025-era setup and worth a second look.
3. Read the consent screen
This is the only moment you see the scopes. Grant the narrow ones. Revoking later stops the very next tool call, from either side.
4. Confirm the tools arrived
Ask "what can this connector do?" Claude lists what it can see. An empty list after a successful sign-in means the token is not being accepted — disconnect and reconnect, which clears a stale grant.
Three rules before you connect twelve of them
- Connect three, not twelve. Every connector adds tool descriptions to every turn. Past a few dozen tools the model starts choosing badly, and the symptom looks like the model being stupid rather than the setup being crowded.
- Know which ones can write. MCP has no approval step of its own (the specification does not define one). A connector that stages changes for you to click chose to.
- Own the list at the org level. On Team and Enterprise, connectors are added once by an Owner and admins can disable specific tool calls — which is the control you want before anyone connects a billing system (Anthropic).
If you only add one and it has to touch paid media, start at the Claude setup guide — then ask it where last month's budget went.
Claude connector questions
What are Claude connectors?
Claude connectors are integrations that let Claude read and act in other software during a conversation. Each one is a remote MCP server: you authorise it with OAuth, and Claude inherits the permissions you already have in that tool rather than getting a copy of your password.
How many connectors can I add to Claude?
Custom connectors are available on Free, Pro, Max, Team and Enterprise, and Free is limited to one custom connector. On Team and Enterprise only an Owner adds a connector for the organisation; members then connect their own account to it.
What is the difference between a Claude connector and an MCP server?
None, technically. "Connector" is Anthropic’s product word for a remote MCP server that Claude can reach over HTTPS. Everything in the connectors directory is an MCP server, and any remote MCP server can be added as a custom connector even if it is not in the directory.
Are Claude connectors free?
Adding one is. What it reaches usually is not: Semrush needs a Semrush subscription, HubSpot needs a HubSpot account, and Zapier spends two tasks from your plan’s quota on every tool call. The connector is a door, not a licence.
Which Claude connector is best for marketing?
It depends on the job. For paid media and SEO execution across Google, Meta, LinkedIn, TikTok, Reddit, GA4 and Search Console, Sprites is the one built for it. For keyword and competitor data, Semrush. For CRM and lifecycle, HubSpot. Most marketing teams end up with three or four, not twelve.
Can a Claude connector change things, or only read?
That is a property of the server, not of Claude. Some are read-only by design, such as Google’s Analytics server. Others can write. MCP itself has no approval step, so a connector that stages changes for you to click is making a deliberate product choice.
Are Claude connectors safe?
Treat one like installing an app with account access. Anthropic’s own documentation says to "only connect to trusted servers" and warns that "malicious MCP servers may include hidden instructions" — the model reads tool descriptions as guidance, so a hostile server can put text there. Check who publishes it, what scopes it asks for, and whether you can revoke it.
Is there an official Google Ads connector for Claude?
No. Google publishes an MCP server for Analytics and nothing for Google Ads, Meta Ads or Search Console. Everything in the paid-media category is either a community wrapper you run yourself or a hosted product such as Sprites.