MCP by the Numbers: What Search Data Reveals About the Ecosystem
Everyone agrees MCP is having a moment. But “a moment” is hard to size. So we ran the ecosystem through Ahrefs — real search volumes, difficulty, and competitor traffic — to see what people are actually typing. The picture is clearer, and more interesting, than the hype suggests.
The category terms are enormous. “mcp server” alone pulls 62,000 US searches a month and 219,000 globally — for an acronym almost nobody had heard of two years ago. But look at the difficulty column: these are 70–80+ out of 100. For a young site, ranking here is a multi-year fight against docs and Wikipedia-tier authority.
| Keyword | US/mo | Global/mo | Difficulty |
|---|---|---|---|
| mcp server | 62,000 | 219,000 | 81 |
| model context protocol | 20,000 | 59,000 | 83 |
| mcp servers | 11,000 | 38,000 | 74 |
| what is mcp | 11,000 | 30,000 | 72 |
Here’s the finding that matters. People don’t just search “mcp server” — they search for specific ones. “playwright mcp” does 14,000 a month. “claude mcp,” “context7 mcp,” “serena mcp,” “supabase mcp,” “zapier mcp” — each is a real, recurring query. And the difficulty is a fraction of the head terms (“mcp inspector” is a 4). This long tail is the ecosystem’s search map: hundreds of named servers, each with its own demand.
| Keyword | US/mo | Difficulty |
|---|---|---|
| playwright mcp | 14,000 | 30 |
| claude mcp | 6,400 | 24 |
| context7 mcp | 5,600 | 31 |
| mcp inspector | 4,100 | 4 |
| serena mcp | 3,700 | 28 |
| github mcp server | 2,500 | 34 |
| zapier mcp | 2,300 | 15 |
| supabase mcp | 1,600 | 22 |
| n8n mcp | 1,500 | 30 |
| postgres mcp | 1,000 | 21 |
A slice of the long tail — there are hundreds more. This is why a page per server matters.
Buyer intent lives in the “best” and “awesome” queries. Volumes look modest, but the traffic potential — how much the top-ranking page actually pulls from the whole cluster — is enormous: ~9,400 for “best mcp servers,” and “awesome mcp servers” sits at a difficulty of 9. This is the most winnable high-value real estate in the space.
| Keyword | US/mo | Difficulty | Traffic potential |
|---|---|---|---|
| best mcp servers | 1,100 | 24 | 9,400 |
| awesome mcp servers | 800 | 9 | 9,300 |
| awesome mcp | 900 | 24 | 6,000 |
The most striking number isn’t search volume — it’s how little the incumbents capture. The current directories rank for a handful of keywords each and pull under 10,000 US visits between them. Nobody owns this category yet.
| Directory | Organic keywords | US traffic/mo | In top 3 |
|---|---|---|---|
| mcp.so | 111 | ~9,200 | 12 |
| glama.ai | 45 | ~3,500 | 2 |
| smithery.ai | 168 | ~1,800 | 13 |
For context: mature SaaS directories rank for tens of thousands of keywords. This is a category in its first innings.
- The demand is real, not hype. Hundreds of thousands of monthly searches, and the per-server long tail keeps growing as new servers ship.
- Discovery is per-server and per-capability. People arrive knowing what they want (“supabase mcp,” “best mcp for scraping”), not browsing a homepage. Structure beats a single big list.
- Naming is distribution. If you ship a server, its name is a keyword. “{tool} mcp” is how people will look for you.
- It’s early. The winnable, high-intent terms are wide open — for now.
We built MCPExplorer to map exactly this — every server, verified, searchable by capability.
Methodology: keyword volumes, difficulty (0–100), and traffic potential from Ahrefs; competitor figures from Ahrefs Site Explorer. United States, July 2026. Search volumes are estimates and a snapshot in time.