AI agents need real search numbers, not guesses
Every agent has opinions about what people search for. seodata.dev gives the numbers instead. The service returns monthly search volume, cost per click, and competition data for any keyword across more than fifty countries. No key is needed for twenty free lookups each day.
One keyword, many reliable sources
Every request routes to the most reliable source available, including Google Ads data, DataForSEO and SerpApi. One consistent response shape returns regardless of which source powers the query. New data types switch on as they ship with nothing to change on the caller's side. The service caches what changes slowly like volume and CPC so nobody pays twice for a keyword someone already looked up.
Free start, clear limits, simple upgrade
Twenty lookups a day are free with no sign-up. Register an email for five hundred a month, or upgrade to unlimited for four dollars ninety-nine a month with no contract. The API returns HTTP 429 until the limit resets. Register, sign in through OAuth, or upgrade to lift the restriction.
How the data breaks down: volume, CPC, competition
The API returns the keyword, monthly search volume, CPC in USD, competition from zero to one, the country, and a list of similar keywords with their own volumes. More than fifty countries are available by two-letter ISO code including US, GB, DE, FR, ES, IT, BR, IN, JP, AU, CA and many others. The full list appears in llms.txt. The MCP tool lookup_keywords accepts up to twenty-five keywords in one call; each keyword counts as one lookup.
Popular keywords that surprise even confident agents
Real numbers often contradict agent assumptions. "Headless cms" gets four thousand four hundred searches a month in the US with CPC twenty-five dollars twenty-four and competition zero point six two. The question forms add up to two thousand twenty searches a month with far easier competition. "Ai code editor" gets two thousand nine hundred a month while "ai powered code editor" gets fourteen thousand eight hundred — same product, five times the demand. "Meal prep app" only gets thirteen hundred searches but "food planner app" and "apps for meal planning" each get nine thousand nine hundred.
Setup that an agent can finish alone
No dashboard required. Plain-text documentation at llms.txt lets an agent finish the sign-up flow independently. Point at seodata.dev/llms.txt and the flow can ask for the six-digit code from the inbox. Add https://mcp.seodata.dev/mcp as a remote MCP server in the client; the setup guide covers Claude, ChatGPT, Meta Muse and others. Anything that can make an HTTP request or speak MCP can use the service: Claude, Claude Code, ChatGPT, Codex, Gemini, Cursor or your own scripts.
What this changes for agent-driven decisions
Models are confident about search demand and often wrong. Real numbers settle it. Agents can stop guessing and start using data before deciding anything: what to write, how to budget ads, which markets appear tiny but aren't, and where expensive clicks hide behind low competition. The one API for SEO data, built for AI agents and the people who run them, returns the numbers every agent needs.