MCP Servers List: Categories, Labels and Where to Look

A good MCP servers list is organised by what you want done, labelled by trust level, and honest about what it does not know. The largest public directory we reviewed for this guide sorts servers into 18 categories, from Sales and SEO to Databases and Security, and shows a status badge and tool count on each card.

How the main lists differ

You will run into three kinds of lists. They answer different questions, so it helps to know which one you are reading.

Type of list	Strength	Weakness
Vendor-backed directory (for example BusinessMCP)	Categories, labels, tool counts, setup steps	Reflects one publisher’s criteria
Official registry	Authoritative about what is registered	Sparse guidance on choosing
Community “awesome” lists on GitHub	Broad and fast-moving	Uneven vetting, many stale links

None of them is a safety certificate. Use them to find candidates, then check each candidate yourself.

The 18 categories on BusinessMCP

According to its directory page, the categories are Sales, Marketing, SEO, Databases, Browser automation, DevOps, Finance, Payments, Dev Tools, Analytics, Communication, Free Growth Tools, AI and ML, Productivity, Research, Design, Security and Writing. Each has a “best of” page, for example best database servers and best SEO servers.

A practical way to use that structure is to pick two categories that match your week. A developer might take Dev Tools and Databases. A marketer might take SEO and Analytics. We cover three of these in depth: developer tools, databases and SEO.

Key takeaway: browse by category, not by popularity. A category list tells you what exists for your task. A popularity list tells you what other people happened to install.

Reading the labels

The directory’s category pages describe their own ranking rules, and they are consistent across the pages we read:

1. Official connectors come first, meaning published by the vendor or hosted by BusinessMCP.
2. Verified entries come next. Verified means the directory holds the server’s own repository or the vendor’s documentation and confirmed that the link resolves. A listing on another directory does not count.
3. Everything else is ordered by number of agent-callable tools.
4. Archived servers are marked as such.

Two details deserve attention. First, the pages say they do not publish install counts for third-party servers, so you cannot infer popularity from position. Second, a dash in the Tools column means the tools have not been listed yet, not that the server has none. More on interpreting badges in official vs verified MCP servers.

What is on a single listing

A listing is where the useful detail lives. Take the Stripe listing. It names the publisher and status, lists four tools by name, gives setup instructions, recommends a restricted read-only key, and states that there are no refund, payout or subscription-change tools. It also offers example prompts, such as asking for failed charges from the past week.

That structure is worth looking for on any list:

• Publisher and status
• Named tools, with a count
• Setup steps and the auth method
• Stated limits, meaning what the server cannot do
• Example prompts that show the intended use

If a list gives you only a name and a one-line blurb, treat it as a starting point for research rather than a recommendation.

Building your own shortlist

Here is a quick method that takes about twenty minutes.

1. Write down three tasks you want an assistant to do this month.
2. Match each task to a category.
3. Open the category page and filter out Archived entries.
4. Keep Official and Verified entries first, then read two or three unlabeled ones if the gap demands it.
5. Open each candidate’s listing and read the tool names and auth method.
6. Discard anything that needs a full-access credential when a restricted one would do.
7. Install one server, test it on low-risk data and only then add another.

Remember that more servers is not better. Every connected tool adds description text to the model’s context and widens the set of actions it might take.

Lists tied to your client

Compatibility matters as much as category. BusinessMCP also publishes client pages, including Claude Code, Cursor and ChatGPT, that describe how to connect. Our own walkthroughs cover Claude, Claude Code and Cursor.

Keeping a list useful over time

Servers get archived, renamed and replaced. Directory pages carry dates (the Dev Tools page we read was updated in September 2026), so look for one. Review your installed set every few months and remove what you no longer use. If you are unsure what an MCP server is in the first place, begin with what MCP servers are.

Sample of what each category contains

To make the categories concrete, here is a small sample taken from the category pages we read. Counts and labels are the directory’s own and change over time.

Category	Entries named on the page	Labels shown
Dev Tools	GitHub, Vercel, Git, Filesystem, GitLab, Netlify, Railway	Official, Verified; Puppeteer marked Archived
Databases	MongoDB, Supabase, Neon, Prisma, Chroma, Pinecone	Several Verified; PostgreSQL, SQLite, Redis shown Archived
SEO	Google Search Console, Google Analytics, Firecrawl, Tavily, Brave Search	Official and Verified near the top

Notice the Archived entries. A directory that shows you what is no longer maintained is more useful than one that hides it, because stale servers are a common source of broken installs and unpatched dependencies.

Common mistakes when using a list

• Treating position as popularity. Order follows labels and tool counts, not usage.
• Installing everything in a category. Pick one per job. Two web fetchers confuse the model.
• Ignoring the auth line. A server that wants a full secret key where a restricted key would do is a warning sign.
• Skipping the date. Dated pages let you judge freshness.
• Trusting a single source. Cross-check a server’s repository or vendor docs yourself.

When a list is not enough

Some needs are too specific for a directory: an internal system, a niche vendor, or a custom workflow. In that case you build a small server of your own or wrap an existing API. Understand the basics first with what MCP servers are, and consider how a server differs from the API underneath in MCP server vs API.

For ordinary needs, though, start from a labelled list, shortlist two candidates per task, test them on harmless data and keep only what earns its place. That discipline matters more than which list you start from.