Standby MCP Server for AI Agents: Multi-Engine Keyword & Autocomplete

An always-on Model Context Protocol (MCP) server and batch Actor that researches live autocomplete keyword suggestions for AI agents (Claude Desktop, Cursor, Windsurf, OpenAI Agents, LangChain) and SEO workflows. It queries five public suggestion endpoints, merges the results and scores every keyword with a transparent demand score and a rule-based search intent.

No proxies, no logins, no cookies, no personal data: only public autocomplete endpoints are called.

Engines

Engine	What you get
google	Suggestions with google_suggest_relevance and suggest type
youtube	Video-search suggestions with relevance
amazon	Buyer queries (marketplace chosen from country; unsupported countries are skipped with an error note)
bing	Web suggestions
duckduckgo	Web suggestions

Expansion modes (expansionMode)

• none: only the seed keyword.
• questions: how, what, why, best prefixes and vs, for suffixes.
• commercial: best prefix and pricing, alternative, review, vs, software, tool suffixes.
• alphabet: seed followed by each letter a to z (27 prompts per seed).

Output fields

uid, keyword, seed_keyword, expansion_prompt, engine, found_in_engines, cross_engine_count, best_rank, google_suggest_relevance (integer or null), demand_score (0 to 100), search_intent, word_count, is_long_tail (4 or more words), suggest_type, country, language.

Search intent

Rule-based, first match wins: transactional (buy, price, deal... or any Amazon suggestion), commercial (best, review, alternative...), comparison (vs, versus, compare...), informational (how, what, why, guide...), navigational (login, official, domain names or NAVIGATION suggestions). Everything else falls back to informational. Use intentFilter to keep only some intents.

Demand score

A relative autocomplete-prominence signal, not a search volume. Computed with exact rational arithmetic and rounded half up:

rank_c = max(0, 11 - max(1, best_rank)) / 10
cons_c = (min(max(cross_engine_count, 1), 5) - 1) / 4
rel_c  = clamp((google_suggest_relevance - 500) / 800, 0, 1)   (rank_c when relevance is null)
demand_score = round(100 * (0.45 * rank_c + 0.30 * cons_c + 0.25 * rel_c))

Results are ranked by demand_score (descending), then best rank, cross-engine count and keyword.

Input

Field	Type	Default
seedKeywords	string[]	["mcp server", "ai agent"] (max 50)
engines	string[]	all five
expansionMode	none, questions, commercial, alphabet	none
country	2-letter code	US
language	2-letter code	en
minWordCount	integer 1 to 20	1
intentFilter	string[]	all intents
deduplicateAcrossEngines	boolean	true
maxItems	integer 0 to 10000	100 (30 in MCP tools)
maxTotalChargeUsd	number or null	no cap

Invalid input is rejected with INVALID_INPUT_ZERO_CHARGE before any network call or charge.

MCP usage

Standby URL: https://<user>--keyword-autocomplete-mcp.apify.actor/mcp (authenticate with your Apify API token).

{
  "mcpServers": {
    "keyword-research": {
      "url": "https://<user>--keyword-autocomplete-mcp.apify.actor/mcp",
      "headers": { "Authorization": "Bearer <APIFY_API_TOKEN>" }
    }
  }
}

Tools:

• research_keywords: same arguments as the batch input (camelCase, maxItems defaults to 30). Returns keywords, count, charged, errors.
• compare_engines: seedKeywords (1 to 5) compared across engines without expansion. Returns per-engine rankings and the overlap of keywords returned by two or more engines.
• list_supported_engines_and_markets: free; lists engines, expansion modes, intents, Amazon marketplaces and pricing.

{"jsonrpc":"2.0","id":1,"method":"tools/call","params":{"name":"research_keywords","arguments":{"seedKeywords":["mcp server"],"expansionMode":"questions","maxItems":20}}}

GET /health returns {"status":"ok"}.

Pricing

Pay per event: $0.001 per keyword returned (keyword-returned); the actor-start event is free. maxItems and maxTotalChargeUsd are hard caps enforced in code, so a run never charges more than you configure. Invalid input and empty results are never charged.

Notes

• Autocomplete reflects what each engine shows for the chosen country and language at request time and can change between runs.
• Failed engine requests are retried (429/5xx) and reported in errors without failing the run.

Related MCP Servers by Mutualistic Labs

Actor	Covers	Store	Standby MCP URL
Multi-ATS Career Jobs Search MCP	Greenhouse, Lever, Ashby, SmartRecruiters, Workable	https://apify.com/mutualistic_housing/multi-ats-career-jobs-mcp	https://mutualistic-housing--multi-ats-career-jobs-mcp.apify.actor/mcp
Google Ads Transparency & Ad Spy MCP	SearchCreatives, SearchSuggestions, content.js ad-copy & video metadata	https://apify.com/mutualistic_housing/google-ads-transparency-mcp	https://mutualistic-housing--google-ads-transparency-mcp.apify.actor/mcp
Multi-Engine Keyword & Autocomplete MCP	Google, YouTube, Amazon, Bing, DuckDuckGo, Brave + google:suggestrelevance & search intent	https://apify.com/mutualistic_housing/keyword-autocomplete-mcp	https://mutualistic-housing--keyword-autocomplete-mcp.apify.actor/mcp
Discourse Forum Search & Threads MCP	community.openai.com, discuss.huggingface.co, meta.discourse.org + any public Discourse	https://apify.com/mutualistic_housing/discourse-forum-mcp	https://mutualistic-housing--discourse-forum-mcp.apify.actor/mcp
Google Trends Realtime & Entities MCP	i0OFE realtime trends, trending/rss news, Knowledge Graph autocomplete entities	https://apify.com/mutualistic_housing/google-trends-realtime-mcp	https://mutualistic-housing--google-trends-realtime-mcp.apify.actor/mcp