Let’s talk
ai era strategy

Best Claude Connectors: The 7 We Use to Run Our SEO (2026)

The best Claude connectors are the ones that close a loop you run every week. Our SEO project ships through seven of them: GitHub, Vercel, a custom SEO connector, Google Sheets, Firecrawl, Gmail and Kit. Here is what each one does for us, how to add your own, and what broke.

Adam Sandler7 min readUpdated
Best Claude Connectors: The 7 We Use to Run Our SEO (2026)

The best Claude connectors are the ones that close a loop you run every week. Our SEO project for this site runs on seven of them, and two do most of the work: GitHub, where every change ships as a pull request, and Vercel, where we confirm it went live. The project has shipped 38 posts and refreshes that way since October 16. Product facts below come from Anthropic's connectors guide, read November 20, 2026.

What are Claude connectors?

Connectors let Claude read your data in other apps and take actions there, such as opening a pull request, searching a Drive or drafting an email. Most are remote MCP servers. Claude inherits your permissions in each app, so a connector can never reach a file, channel or record you could not open yourself.

You find them in the Connectors Directory, where each one has a page listing what it can read and write. Web connectors work on claude.ai, Cowork, Claude Desktop and the mobile apps. Desktop extensions are a separate kind that runs only in Claude Desktop.

Which Claude connectors are best?

The best connectors are the ones that close a loop you run every week. For us that is GitHub, Vercel, a custom SEO connector, Google Sheets, Firecrawl, Gmail and Kit. Pick by the job: one connector that ships work beats ten you only read from.

Here is the list we actually run, in order of how much work each one carries.

1. GitHub: where the work ships

Every post, refresh, schema fix and report change goes out as a pull request. The thread writes the change, opens the PR, reads the review bots' comments, fixes what they find and merges once checks pass. If you only add one connector for real work, add this one.

2. Vercel: proof it went live

A merge is not a launch. The Vercel connector tells us when the production deploy reads READY and fetches the live page, so a thread can check the new section is really there before it reports back. On October 22 it showed a failed production build that the merged PR alone did not.

3. A custom SEO connector: our data

Our keyword and Search Console tool, OpenSEO, is self-hosted. We reach it through a custom connector for Google Search Console, GA4, URL inspection and keyword research. Paid lookups run against a $15 research budget that we log line by line.

4. Google Sheets: the owner's research

The project's strategy started from a keyword sheet the owner built by hand. The Sheets connector let the first thread read it directly, so nobody pasted data into a chat.

5. Firecrawl: sourcing facts

Every product claim in a post needs a source. Firecrawl searches and reads the pages, including the help center article this post cites, and returns clean text a thread can quote from.

6. Gmail: drafts, never sends

Backlink outreach goes through Gmail as drafts only. The owner reads and sends every email. That one rule is why we are comfortable leaving this connector on.

7. Kit: did anyone sign up?

Traffic means little without signups. The Kit connector shows which forms grew the list, which is how we found that the site's own forms had added no one in eight days while a Kit landing page carried all the list growth.

How do you add a connector to Claude?

Go to Customize, then Connectors, at claude.ai, press the plus button and pick a service from the directory. Sign in to that service when asked and review what it can access. On Team and Enterprise plans an Owner turns the connector on for the organization first, then each person signs in.

Once a connector is on, Claude can bring it into a conversation when it fits the request; you do not have to name it each time. In chat you can still switch connectors on or off per conversation from the plus menu.

What is a custom Claude connector?

A custom connector is any remote MCP server you add by its URL instead of picking it from the directory. Claude reaches it from Anthropic's cloud, not from your computer, so the server must be reachable over the public internet. Free plans get one custom connector; paid plans can add more.

To add one, open Customize, then Connectors, choose Add, then Custom, then Web, and paste the server's URL. You pick how people sign in: OAuth, a fixed header such as an API key, or no sign-in. Our SEO connector taught us the catch: it points at a tool on one Mac. When that machine is unreachable, the connector returns errors, and on those days a thread runs on the Mac through Claude Code Remote Control instead. Our November 9 to 15 report came in that way.

Do Claude connectors work in Claude Code?

Yes. Anthropic's docs list Claude, Claude Desktop, Claude Code and the API as places connectors work. In a Claude Code project, every cloud thread starts with your claude.ai connectors, so a thread can open a pull request or pull Search Console data without any setup inside the thread.

That is most of why a Claude Code project works for us. The coordinator starts a thread per task, and each thread arrives with GitHub, Vercel and the SEO data already connected. If you are deciding between Claude's two work modes, our Cowork vs Claude Code comparison covers where connectors fit in each.

Are Claude connectors safe?

They are as safe as the permissions you grant. Directory connectors are reviewed; custom ones are not, so only add servers from people you trust. On Team and Enterprise plans an Owner can set each tool to always allow, needs approval or blocked, for example reading email but never sending it.

Our rule is simpler than any setting: give write access only where a person still approves the result. GitHub writes pull requests, but review and checks gate the merge. Gmail writes drafts, but a person sends them.

How many connectors should you turn on?

Fewer than you think. Each connector adds tools Claude has to consider. Anthropic suggests switching Tool access to "On demand" once you have 10 or more active, so tools load only when needed.

What did running a project on connectors teach us?

  • Write connectors earn their place; read connectors are optional. GitHub and Vercel finish work. Most others feed it, and a thread can often work around a missing one.
  • A custom connector is only as available as its server. Ours fails when one computer is asleep. Plan a fallback before you depend on it.
  • Data should arrive by connector, not by paste. The owner's standing rule is that no thread ever asks for data to be copied in. Connectors are how that rule holds.
  • Budget the paid ones. Research calls cost money per lookup. A written budget and a ledger keep our recurring checks under a dollar a month.

Where to start

Add GitHub and one data source you check every week, then give Claude a task that needs both. If you want to see connectors doing site work, our guide to auditing any website with Claude and MCP is a good first run. If you run Claude Code for clients and want to package that into a service, the Claude Code Service OS covers the offer, pricing and outreach.


Connectors work better when Claude already knows your brand. In Brand Architect, Ophelia builds a knowledge base from your website or materials that any AI tool can read. Build your knowledge base →