Find similar websites with an AI agent and one API call

If you want to find similar websites with AI, the useful setup is not a chat window but an agent that calls a data tool. A language model can guess which sites resemble each other. A keyword overlap lookup can show it. Wire the second into your coding agent and you can ask for "sites like these twelve domains" and get a table back without leaving the editor.

This guide connects seogeoaeo.ai's Competitor Finder to an agent over MCP or REST, caps what each run can cost, and loops it over a list of domains.

What the tool returns #

Competitor Finder takes one public URL and an optional market code. It reduces the URL to its domain, compares that domain's Google keywords with other sites, and returns up to 30 domains that share the most. Each row carries the shared keyword count, the other site's total keywords, the share of its keywords that overlap, estimated monthly visits and average rank. The input domain's own keyword count and visits come back too, for scale.

Sites where at least 5% of their keywords overlap are flagged as close competitors. Very large general sites are kept separate so they do not dominate the list.

Market codes cover 20 countries: us, gb, ca, au, ie, in, sg, nz, za, de, fr, es, it, nl, se, pl, br, mx, jp and ae.

Price: 38 credits per run.

Connect over MCP #

Create a key at seogeoaeo.ai/api-keys. Keys start with sga_ and are shown once, so put yours in an environment variable rather than a file you commit.

For Claude Code:

claude mcp add --transport http --scope user seogeoaeo https://seogeoaeo.ai/api/mcp --header "Authorization: Bearer $SEOGEOAEO_API_KEY"

Cursor and other clients that read an mcpServers file take the same URL and header. Every tool appears under its slug, so this one is competitor-finder.

Or call REST directly #

curl -X POST https://seogeoaeo.ai/api/v1/tools/competitor-finder/runs \
  -H "Authorization: Bearer $SEOGEOAEO_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{"input":{"url":"https://example.com","market":"gb"}}'

Guard the spend #

Two arguments keep an agent loop from surprising you.

• maxCredits makes the API refuse any run priced above the number you set. Pass 40 for this tool and a mistaken call to a pricier tool fails instead of charging.
• idempotencyKey lets the agent retry after a timeout without paying twice for the same run.

check-credits is free, so the agent can read the balance before a batch and stop if it is low. Runs that fail on the platform side are refunded. One thing to plan for: a lookup that finds the domain but no overlapping sites is still charged, because the data provider bills for it. A domain the provider has no record of is refunded.

A loop that works #

Put the domains in a text file and give the agent a prompt along these lines:

"For each domain in seeds.txt, call competitor-finder with market us and maxCredits 40. Keep only close competitors. Write one CSV with columns seed, similar_domain, shared_keywords, overlap_share, est_visits. Then list any domain that appears as a close match for three or more seeds."

The last line is where the value is. A domain that keeps showing up across several seeds is a central player in that niche. Twelve seeds cost 456 credits, well inside the 2,000 on the $29 Indie plan. Agent runs cost the same as browser runs.

Chain the next call #

Once the agent has close matches, it can go one step further on the interesting ones:

• competitor-keywords (35 credits) for a site's top 100 keywords by estimated traffic.
• keyword-gap (85 credits) for the top-20 keywords a match ranks for that a given site misses.
• backlink-finder (70 credits) for the 100 strongest sites linking to a match.

Where a model alone still helps #

Keyword overlap only sees Google search. Brands that compete through sales teams or paid ads will be missing. For those, asking a chat model "what products are like X" fills gaps, as long as you ask more than once and keep only names that repeat. Treat that as a second column, not a replacement.

Limits #

The lookup covers a whole domain and one country per run. Visit and keyword figures are estimates from a search data provider. There is no rank tracking, so a job that needs daily positions belongs to a tracking suite.

Full tool schemas are in the seogeoaeo.ai agent reference, and the tool page is at seogeoaeo.ai/tools/competitor-finder.