Ask an AI assistant "where does my site rank for best crm for startups?" and it will either guess or refuse. LLMs don't have search volumes, today's Google results or backlink counts. They make them up. The fix is to give the agent tools that fetch real data, and the Model Context Protocol (MCP) is now the standard way to plug those tools into Claude, Cursor, VS Code and agent frameworks.
This post covers:
| Tool | What the agent gets |
|---|---|
keyword_metrics |
Google search volume, CPC, competition, keyword difficulty, intent and 12-month trend for up to 100 keywords |
google_serp |
Live Google results (top 10-100) for any country and device, your domain's position, AI Overview text and sources, People Also Ask |
domain_overview |
Organic traffic estimate, ranked-keyword count, backlinks and referring domains for up to 100 domains |
ranked_keywords |
Up to 100 keywords a domain ranks for, with position, volume and URL |
backlinks |
Up to 100 backlinks or referring domains for a domain or page |
google_trends |
Interest over time and by region for up to 5 terms |
youtube_search |
Ranked YouTube results with views, channel and date |
youtube_video |
Likes, comments, subscriber count, tags, plus the transcript if you ask |
All tools are read-only. There are no logins, cookies or proxies on your side: the data comes from licensed SEO data feeds.
{
"mcpServers": {
"seo-data": {
"url": "https://steady-scrape-1--seo-data-mcp-server.apify.actor/mcp",
"headers": { "Authorization": "Bearer YOUR_APIFY_TOKEN" }
}
}
}
That's it. The server runs in Apify Standby mode: it starts on the first request (a few seconds of cold start in our tests) and stays warm while you work.
Then just ask:
import asyncio, os, httpx2
from mcp import ClientSession
from mcp.client.streamable_http import streamable_http_client
URL = "https://steady-scrape-1--seo-data-mcp-server.apify.actor/mcp"
async def main():
headers = {"Authorization": f"Bearer {os.environ['APIFY_TOKEN']}"}
async with httpx2.AsyncClient(headers=headers, timeout=60) as hc:
async with streamable_http_client(URL, http_client=hc) as (r, w, *_):
async with ClientSession(r, w) as s:
await s.initialize()
res = await s.call_tool("google_serp", {"keyword": "best crm for startups",
"target_domain": "hubspot.com"})
print(res.structured_content["targetPosition"])
asyncio.run(main())
In our tests most tools answered in 1-5 seconds. A top-100 SERP takes about 15 s and Google Trends about 14 s, because both are fetched live.
You pay per successful call. Errors, invalid arguments and empty results are free, and there's no compute charge on top.
| Session | Cost (Free plan) |
|---|---|
| Rank check: 5 keywords, top 10 each | $0.035 |
| Keyword research: 100 keywords | $0.395 |
| Competitor audit: 10 domains + 2 ranked-keyword lists + 1 backlink list | $0.340 |
| YouTube research: 2 searches + 5 videos, 3 with transcripts | $0.161 |
Per tool: $0.006 per 10 SERP results, $0.039 per 10 keywords, $0.111 per 10 domains, $0.071 per ranked-keywords call, $0.082 per backlinks call, $0.035 per Trends call, $0.006 per 20 YouTube results, $0.018 per video (+$0.018 for a transcript), and $0.005 once per session. Paid Apify plans get 5-15% off the tool prices.
Agents can loop, so the server has guardrails. It refuses calls that would push you past the maximum charge you set for a session, before fetching anything. It also limits a session to 60 calls a minute.
Disclosure: we (SteadyScrape) built this server and get paid when it's used.
SEO MCP Server: Keyword, SERP & Backlink Data (steady-scrape/seo-data-mcp-server) is on the Apify Store. Paste the config above, ask your agent an SEO question, and check the billing field in each response to see exactly what was charged.
No MCP client yet? Run the Actor normally with input like {"calls": [{"tool": "google_serp", "arguments": {"keyword": "seo mcp server"}}]} and each call becomes one dataset row. Our post-publish test of exactly that took one SERP page plus the session fee, about $0.011.