| name | seo-keyword-research |
| description | Use when researching keywords for a website domain, finding related or long-tail terms, grouping intent, scoring opportunities, calculating a 0-100 Keyword Score, or producing a Markdown report. |
| compatibility | Requires the official DataForSEO MCP server with the DATAFORSEO_LABS module enabled and filesystem write access. |
Keyword Research and Opportunity Scoring
Follow the official Related Keywords, Suggestions, Ranked Keywords, and Bulk KD documentation.
Input
- Require a project domain; if absent, ask and wait, even with a seed. Use a supplied seed for seed mode; otherwise use domain mode. With both, default to seed unless domain mode is explicit.
- Default to
United States/en; honor overrides and disclose scope.
- Normalize to lowercase hostname without credentials, port, suffix, trailing dot, or leading
www.. Reject malformed input.
DataForSEO MCP workflow
The request authorizes these billable calls only. Ask before retries, pagination, or extras.
- Seed: call
dataforseo_labs_google_related_keywords with limit: 200, then dataforseo_labs_google_keyword_suggestions with limit: 100.
- Domain: call
dataforseo_labs_google_ranked_keywords with the normalized target, limit: 100, item_types: ["organic"], and search-volume-descending order.
- Deduplicate case-insensitively, preserve sources, and call
dataforseo_labs_bulk_keyword_difficulty once for up to 1,000 unique keywords; merge KD by normalized keyword.
- Validate statuses. Extract keyword, volume, CPC, Ads competition/level, KD, intent, and rank/URL. Never invent metrics; state coverage.
Cost accounting
Log each call's endpoint and top-level cost USD. Sum unrounded values. Scope: Total cost: x,xx USD (decimal comma, two digits). Include zero; missing cost means incomplete subtotal; name affected calls.
Analysis
Classify one intent in order: URL/domain or brand navigation -> Navigational; buy, price, deal, near me, or brand+product -> Transactional; best, review, comparison, or vs -> Commercial; how to, what is, guide, or tutorial -> Informational. Otherwise use main_intent, then Informational (inferred).
With both metrics calculate:
opportunity = search_volume / (keyword_difficulty + 10)
Sort descending and surface 20. Never coerce missing values to zero.
For Keyword Score, let K be up to 50 top opportunity rows. Show each component:
- volume =
30 * min(1, log10(1 + average_search_volume(K)) / 5);
- low difficulty =
25 * count(KD < 30) / count(valid KD in K);
- intent diversity =
15 * distinct_intent_buckets(K) / 4;
- CPC =
15 * min(1, log10(1 + average_CPC_USD(K)) / log10(11));
- long-tail =
15 * min(1, breadth / 100), where breadth is suggestions in seed mode or ranked keywords of at least four words in domain mode (label as proxy).
Sum without reweighting, clamp 0-100, and round to one decimal. Score unavailable components zero and disclose them. Cite strongest and weakest signals.
Report
Use requested report root or <current-working-directory>/SEO; create its normalized domain child. Sanitize components to letters, numbers, dots, hyphens, and underscores, capped at 140 characters. Write <report-root>/<domain>/<YYYY-MM-DD>_Keyword-Analysis_<target>.md, sanitizing target separately. Start with local ISO date.
Include Scope/total cost; summary; score components; top-20 opportunities; intent groups; related keywords; suggestions/proxy; metrics; content priorities; formulas; call log; timestamp; coverage; limitations; and official links. Return absolute path and summary.
