Most "best MCP server" roundups copy README blurbs without sending a single JSON-RPC packet. On November 20, 2026, we probed 14 live marketing, SEO, analytics, CRM, and web-data MCP servers directly over Streamable HTTP initialize, tools/list, RFC 9728 OAuth protected-resource metadata checks, and GitHub API inspections. Start with Parse MCP for keyless 90-day AI visibility benchmarks, GA4 and GSC MCPs for zero-cost organic analytics, Ahrefs or Semrush if you already hold paid SEO seats, and Firecrawl when an assistant needs clean Markdown from stubborn HTML.
We tested the endpoints and repositories in this guide directly on November 20, 2026, and paired that live engineering benchmark with our 911,413-answer study, How AI recommends MCP and agent-native tools (September 1 to November 19, 2026). Parse tracks AI visibility across 853,414 active brands and 1,774 markets, backed by 744,460 counted-engine answers (378,499 ChatGPT Search and 365,961 Google AI Mode) and 4,053,878 citations across 73,847 buyer questions in our 90-day public MCP snapshot (August 23 to November 20, 2026).
Live marketing, SEO, analytics, ops, and web-data MCP servers verified by JSON-RPC and GitHub checks (Nov 20, 2026)
Jump in ChatGPT Search MCP mention share from Sep (0.660%) to Nov 2026 (1.499%)
Of 6,292 MCP mentions across 911,413 organic AI answers that surfaced unprompted (5,975 of 6,292)
Estimated schema tokens for Parse MCP's 6 unauthenticated tools vs 450 tokens for Cloudflare Docs (2 tools)
Why marketers need an MCP shortlist backed by live probes
MCP adoption inside buyer recommendations is moving faster than explicit MCP search volume. Across 130,776 active organic buyer questions in Parse's corpus, only 18 explicitly named MCP (3,146 asked for MCP, AI agents, Claude/Cursor, headless setups, APIs/SDKs, or CLI workflows). Yet across 911,413 September 1 to November 19, 2026 answers (463,371 ChatGPT Search, 440,174 Google AI Mode, 6,780 Claude reference answers, and 1,088 Perplexity), 6,292 answers mentioned MCP anyway, with 94.96% (5,975 of 6,292) surfacing unprompted.
ChatGPT Search drove the sharpest surge, more than doubling its MCP mention share (+127.12%) from 0.660% in September (1,063 of 161,053 answers) to 1.499% in November (1,711 of 114,137 answers), putting it 2.72× ahead of Google AI Mode's 0.552% (595 of 107,732 answers). Teams wiring marketing workflows into Claude Code, Cursor, or ChatGPT face messy vendor docs, because old guides still list dead URLs and archived repos.
Our November 20 live probes caught three recurring traps:
- Dead docs URLs: Legacy Ahrefs docs (
docs.ahrefs.com/docs/mcp/reference/introduction) return404; canonical docs sit athttps://docs.ahrefs.com/en/mcp. - Broken path guesses:
mcp.hubspot.com/mcpreturns404; HubSpot's endpoint ishttps://mcp.hubspot.com(https://developers.hubspot.com/mcp). - Deprecated local repos:
makenotion/notion-mcp-server(4,719stars) now redirects users to Remote Notion MCP athttps://mcp.notion.com/mcp.
Which of the 14 live-tested MCP servers fits each marketing job?
No single MCP server covers search telemetry, product analytics, web extraction, and sprint ops. Split your stack by job so assistants pull exact missing data without loading unused schemas.
The table below lists all 14 MCP servers probed on November 20, 2026, with verified endpoints, transport and auth models, and pricing.
| Server & category | Verified endpoint or repo | Transport & auth model | Key capabilities & pricing |
|---|---|---|---|
| Parse MCP (AI visibility) | https://agent.parse.gl/mcp (alias mcp.parse.gl/mcp) | Streamable HTTP; keyless public view (5 read tools, 6 listed unauth) + Bearer / OAuth | 16 canonical tools; free 90-day index + weekly check + $5 signup credit; $0.01/live check or fetch; no subscription |
| Ahrefs MCP (SEO & backlinks) | https://api.ahrefs.com/mcp/mcp | Streamable HTTP (?tools= filter); OAuth 2.0 (RFC 9728, apiv3-mcp) | 10 groups (site-explorer-*, keywords-explorer, brand-radar-*, gsc); paid Ahrefs plan (Lite+) + API units |
| Semrush MCP (SEO & traffic) | https://mcp.semrush.com/v2/mcp | Streamable HTTP; OAuth 2.0 (mcp.access) or Bearer API key | Keywords, traffic/backlinks, AI Visibility / Prompt Research, rank tracking, audit; paid Semrush / API plan |
| GA4 MCP (Web analytics) | googleanalytics/google-analytics-mcp (3,838 stars) | Local stdio (pipx analytics-mcp) or HTTP; Google Cloud ADC / read-only OAuth | run_report, run_funnel_report, run_conversions_report, run_realtime_report; free Apache-2.0 |
| GSC Community MCPs (Search) | AminForou/mcp-gsc (2,249 stars); crunchtools/mcp-google-search-console | Local stdio (npx/uvx/Python); Google OAuth 2.0 or service account | 20 tools (get_search_analytics, compare_search_periods, inspect_url_enhanced, manage_sitemaps); MIT / AGPL-3.0 |
| HubSpot MCP (CRM & ops) | https://mcp.hubspot.com | Streamable HTTP + local CLI; OAuth 2.0 (RFC 9728) | Read/write contacts, companies, deals, tickets, engagements; included in HubSpot plans |
| PostHog MCP (Product & LLM) | https://mcp.posthog.com/mcp | Streamable HTTP; OAuth 2.0 or API token (160 scopes verified) | Funnels, cohorts, feature flags, A/B tests, replays, LLM analytics; free tier + usage-based |
| Notion MCP (Docs & wikis) | https://mcp.notion.com/mcp | Streamable HTTP (replaces local repo); OAuth 2.0 | Semantic search, Markdown-native page reads/edits, database queries; included in Notion plans |
| Slack MCP (Team alerts) | https://docs.slack.dev/ai/slack-mcp-server/ | Remote HTTP/SSE; OAuth 2.0 / Dynamic Client Registration | Search messages/files, post updates, canvases, lists; included with admin MCP approval |
| Linear MCP (Marketing ops) | https://mcp.linear.app/mcp | Streamable HTTP; OAuth 2.0 (read, write) | Search, create, and update issues, projects, cycles, docs, comments; included in Linear plans |
| Stripe MCP (Revenue data) | https://mcp.stripe.com | Streamable HTTP + stripe agent setup; OAuth 2.0 or restricted key (mcp) | Customers, invoices, subscriptions, payments, balances, Stripe docs; free with Stripe account |
| Firecrawl MCP (Extraction) | https://mcp.firecrawl.dev/mcp (7,872 stars) | Streamable HTTP + local npx firecrawl-mcp; Firecrawl API key | firecrawl_scrape, firecrawl_map, firecrawl_crawl, firecrawl_search, firecrawl_interact, firecrawl_parse; MIT + credits |
| Playwright MCP (Browser QA) | microsoft/playwright-mcp (39,934 stars) | Local stdio / SSE (npx @playwright/mcp@latest); no external auth | 65+ browser_* tools using accessibility snapshots; free Apache-2.0 |
| Cloudflare Docs MCP (Ref) | https://docs.mcp.cloudflare.com/mcp | Streamable HTTP (SSE); public read-only, zero auth | 2 tools (search_cloudflare_documentation, migrate_pages_to_workers_guide); free public endpoint |
Treat that matrix as an architecture menu rather than a list to install all at once. For Claude Code or Cursor AI visibility workflows, pairing one public visibility server (Parse), one owned analytics source (GA4 or GSC), and one Markdown fetcher (Firecrawl) covers most weekly work.
How auth models, OAuth scopes, and schema bloat affect agent context
Connecting five heavy MCP servers to one coding agent can wreck performance before your first prompt finishes. Two technical constraints decide whether an MCP stack stays fast or stalls out: authentication mechanics and tools/list schema token footprint.
On the auth front, only Parse MCP (public index view) and Cloudflare Docs MCP answered our unauthenticated HTTP POST JSON-RPC initialize + tools/list probe with HTTP 200. Every SaaS ops, CRM, and paid SEO endpoint (Ahrefs, Semrush, HubSpot, PostHog, Notion, Linear, Stripe) returned HTTP 401 advertising RFC 9728 OAuth protected-resource metadata (/.well-known/oauth-protected-resource).
Scope discipline varies wildly across those OAuth servers. Linear MCP keeps its surface crisp with two scopes (read and write). PostHog MCP, by contrast, advertises 160 granular OAuth scopes (insight:read, experiment:write, llm_analytics:read, web_analytics:read) so teams can lock read-only analytics tokens away from billing or feature-flag mutations. Always audit scopes before giving an autonomous agent write access to HubSpot, Notion, or Slack.
Schema size hits even harder on coding agents. We measured the exact tools/list JSON-RPC byte and token footprint (characters / 4) on every unauthenticated endpoint and checked vendor guidance across heavy tool suites.
Minimal remote (Cloudflare Docs MCP). Exposes 2 tools in 1,801 bytes (450 est. tokens). Zero auth and tiny context overhead for docs lookup.
Scoped public analytics (Parse MCP). Unauthenticated tools/list exposes 6 tools (check_brand, resolve_ids, query_metrics, search_answers, search_cited_pages, get_brief) in 21,520 bytes (5,380 est. tokens; OpenAPI is 18,274 bytes).
Heavy browser suite (Playwright MCP). Exposes 65+ browser_* tools. Microsoft's own README warns coding agents to prefer Playwright CLI+SKILLs when large MCP schemas and accessibility trees strain context windows.
Notice how smart servers solve schema sprawl. Ahrefs MCP lets you append ?tools= query parameters to restrict the endpoint to specific tool groups (such as brand-radar-simple or keywords-explorer) rather than loading all 10 groups at once. Google Search Console still has no Google-hosted remote MCP server, so teams choose between the 20-tool MIT AminForou/mcp-gsc server (2,249 stars, updated November 6, 2026, defaulting to read-only OAuth unless GSC_ALLOW_DESTRUCTIVE=true is set) or crunchtools/mcp-google-search-console (AGPL-3.0, also supporting Streamable HTTP container mode). Meanwhile, Firecrawl MCP (7,872 stars) compresses multi-hundred-kilobyte HTML pages into clean Markdown before text ever touches your model context.
How AI models recommend AI visibility and marketing MCP tools today
We make Parse, and we don't hide behind cherry-picked screenshots when evaluating our own category. Inside our 911,413-answer MCP study, we compared 85 standard AI-visibility buyer questions (2,391 answers across ChatGPT Search, Google AI Mode, and Perplexity) against 5 explicit MCP/agentic AI-visibility prompts (158 answers: 79 on ChatGPT Search and 79 on Google AI Mode).
The split between standard buyer prompts and MCP-focused prompts is dramatic, and the differences between ChatGPT Search and Google AI Mode are even sharper.
| Platform | Standard AI visibility (2,391 ans) | ChatGPT Search MCP prompts (79 ans) | ChatGPT #1 wins & avg pos | Google AI Mode MCP prompts (79 ans) |
|---|---|---|---|---|
LLM Pulse (llmpulse.ai) | 1.21% | 84.81% | 11 #1 picks (avg 4.45) | 27.85% (6 #1 picks) |
Peec AI (peec.ai) | 49.85% | 70.89% | 8 #1 picks (avg 4.32) | 46.84% (3 #1 picks) |
Otterly AI (otterly.ai) | 51.07% | 62.03% | 6 #1 picks (avg 4.67) | 45.57% (0 #1 picks) |
SE Ranking (seranking.com) | 10.66% | 60.76% | 2 #1 picks (avg 8.04) | 45.57% (5 #1 picks) |
Ahrefs (ahrefs.com) | 36.89% | 54.43% | 0 #1 picks (avg 7.91) | 10.13% (3 #1 picks) |
Cite42 (mcp.cite42.dev) | <0.2% (<5 ans) | 41.77% | 4 #1 picks (avg 6.42) | 1.27% (1 ans) |
Scrunch (scrunch.com) | 21.83% | 37.97% | 0 #1 picks (avg 6.71) | 5.06% (0 #1 picks) |
| TryProfound / Profound | 40.28% / 11.25% | 32.91% / 17.72% | 1 / 0 #1 picks | 11.39% / 34.18% |
Promptwatch (promptwatch.com) | 6.65% | 17.72% | 3 #1 picks (avg 5.07) | 11.39% (0 #1 picks) |
Parse (parse.gl) | 0.33% | 13.92% | 7 #1 picks (avg 1.82) | 0.00% (0 #1 picks) |
Semrush (semrush.com) | 48.68% | 6.33% | 0 #1 picks (avg 8.80) | 8.86% (3 #1 picks) |
Look closely at what that table teaches both buyers and product marketers:
- Standard prompts stay sticky around early category incumbents. On the
85generic AI-visibility prompts,Otterly AI(51.07%),Peec AI(49.85%),Semrush(48.68%),TryProfound(40.28%), andAhrefs(36.89%) take the lion's share of mentions, while Parse sits at just0.33%(8of2,391). - MCP prompts reward documentation, GitHub, and fast-moving specialists on ChatGPT Search. On the
5MCP-focused AI-visibility prompts on ChatGPT Search,LLM Pulsesurges from0.41%(5of1,208standard ChatGPT answers) to84.81%,Cite42jumps to41.77%, and Parse jumps 42× to13.92%(11of79), winning#1in7of11appearances at a1.82average rank because ChatGPT citesparse.gltechnical docs directly. - Google AI Mode looks elsewhere. On those exact same
79Google AI Mode answers, Parse earned zero mentions, whileRadarkitwon18.99%(9#1picks) andPeec AI,SE Ranking, andOtterly AIled. Why? Across all58,092citation links on MCP-mentioning answers, ChatGPT Search leaned2.4×harder on Docs/Dev pages (30.73%vs12.65%) andgithub.com(4.01%vs2.40%), whereas Google AI Mode cited YouTube (9.32%vs0.00%), Reddit (3.47%vs0.01%), and MCP registries likemcpmarket.comandmcpservers.org(0.79%vs0.02%).
If you want deeper comparison across AI-visibility vendors specifically, pair this 14-server buyer's guide with our seven headless AI visibility platforms comparison, our ChatGPT connectors guide, and our GitHub and DevRel AI visibility playbook.
Which three marketing MCP stacks work without overloading your agent
Instead of wiring 14 servers into one config, run three focused profiles:
- Zero-cost SEO & AI visibility audit: Pair Parse MCP (
https://agent.parse.gl/mcp, keyless public index +$5signup credit), GA4 MCP, and GSC MCP (AminForou/mcp-gsc, read-only OAuth) for 90-day AI answers, organic queries, and conversions without subscriptions. - Technical SEO & Markdown debugging: Pair Parse MCP with Firecrawl MCP (
https://mcp.firecrawl.dev/mcp) to compare AI-cited gaps against live extracted Markdown, using Playwright CLI for JS pricing checks. - Growth ops loop: Pair Parse MCP with Linear MCP (
https://mcp.linear.app/mcp) or Notion MCP (https://mcp.notion.com/mcp) and Slack MCP, adding Ahrefs MCP (?tools=brand-radar-simple,keywords-explorer) or Semrush MCP only when keyword or backlink queries demand it.
Here is a lean .mcp.json starter for Cursor or Claude Code combining Parse, Firecrawl, and GSC:
{
"mcpServers": {
"parse-agent": {
"url": "https://agent.parse.gl/mcp",
"headers": {
"Authorization": "Bearer ${PARSE_API_KEY}"
}
},
"firecrawl": {
"url": "https://mcp.firecrawl.dev/mcp",
"headers": {
"Authorization": "Bearer ${FIRECRAWL_API_KEY}"
}
},
"gsc": {
"command": "uvx",
"args": ["mcp-gsc"]
}
}
}
For quick keyless discovery, omit the Authorization header on parse-agent and check_brand, resolve_ids, query_metrics, search_answers, and search_cited_pages still answer immediately.
Run this workflow with your agent
Before signing an annual contract for any MCP visibility or SEO suite, test your own category inside Claude Code, Cursor, Codex, or ChatGPT using Parse MCP (https://agent.parse.gl/mcp). Reading the 90-day public index and recording notes with log_work is free, and live page inspections (fetch_live_pages) or fresh checks (ask_ai_engines) quote at $0.01 inside a monthly cap you control.
Run get_brief, then check_brand for "<our domain>" and query_metrics over the last 90 days
across ChatGPT Search and Google AI Mode.
Call search_cited_pages and search_answers to list the top 10 third-party pages citing
competitors on our primary buyer topics while skipping us.
If we have GA4 or GSC connected, cross-check organic clicks and AI-referral conversions
on those matching landing pages, then save a prioritized sprint plan with log_work.
Your assistant calls check_brand, query_metrics, search_cited_pages, search_answers, get_brief, and log_work in sequence, storing an audit trail any teammate can reopen with list_work or verify next month via track. Explore market research workflows, competitor comparison workflows, or review transparent pay-as-you-go pricing before adding credits.
Want to see who AI recommends in your market and which sites shape those answers? research your market with Parse.