AI Visibility MCP Servers Compared: Tools, Sign-in and Plan Gates (2026)

The AI visibility MCP server with the most published detail at the lowest entry price is Cituna's: 31 tools, 22 that read your daily tracking of seven AI engines and your Search Console data from the $49 Starter plan, and 9 that write, including running a fresh scan, on the $119 Pro plan, hosted at mcp.cituna.com or run locally (Cituna's MCP page, read 22 October 2026). For other jobs, Cite42 and Manifold run new checks per call with no subscription, LLM Pulse puts about 130 tools on every plan from €49, and GEOly documents up to 53 tools with write access chosen at sign-in (all read 22 October 2026).

The question that sorts these servers, as Cite42's own comparison puts it, is whether the assistant runs a new check right now or reads data a platform has already collected. Most of the 21 servers below read stored data, Cite42 and Manifold run new checks per call, and a few, Cituna among them, read stored data and can also start a fresh run.

How do AI visibility MCP servers compare?

What each server exposes: tools, read and write, hosting and sign-in

The table above is about price. This one is about the server itself, from each vendor's own MCP documentation. "Not stated" means the page we read does not say, not that the feature is missing.

How we chose

This is a desk comparison of published pages. For each server we read the vendor's own MCP documentation and pricing page on the date shown, and we did not install, buy or run these servers for this note; Cite42's and Manifold's facts come from their own comparison pages. A server is numbered if its vendor documents an AI visibility server and a price or plan we could read. Cituna comes first because we build it; judge it by the same tests. The rest follow by the monthly price of the cheapest plan that includes MCP, low to high, with pay-per-call and free-tier servers first and Enterprise-only servers last. Servers inside SEO suites, and trackers that name MCP on a pricing page without documenting it, are covered after the numbered list.

Does the assistant run a new check, or read data already collected?

This is the question that sorts every server here, and it decides what your assistant can honestly tell you.

A stored-data server reads a tracking program: you set up prompts once, the vendor asks the engines on a schedule, and your assistant queries the history, such as trend lines, share of voice and which sources gained or lost. Peec AI, Otterly.AI, LLM Pulse, GEOly, Geoptie, Scrunch, Profound and the suite servers work this way, and only this kind can answer "did we gain ground this month?"

An on-demand server asks the engines when the assistant calls it and returns what they say now. Cite42 and Manifold work this way: no subscription and no history unless a tracker keeps one, but any prompt can be checked without setup, which suits an audit, a pitch or checking a fix the day it ships.

The trade-off is statistical. Engines change their answers from run to run, so one fresh check is one draw: it tells you what an engine said once, not how often it says it. A stored program of daily runs over weeks is what turns draws into a rate. AI visibility tools with a free trial works out how many runs a result needs before it means something.

Cituna's server does both. Its read tools return the daily grid across seven engines and the answer text behind each cell, and on Pro the run_scan tool starts a fresh scan of a tracked brand, using one scan from the plan's quota. The second half of the question, what to do about a gap, is where write tools come in: Cituna's queue_article and set_gap_status, Otterly.AI's prompt and audit tools, Scrunch's prompt and persona edits, GEOly's trigger-monitoring tool and LLM Pulse's recommendation runs all change something on the vendor's side.

The 10 AI visibility MCP servers, one by one

1. Cituna (best for stored daily data plus fresh scans, with Search Console)

Cituna's MCP server has 31 tools (read 22 October 2026). The 22 read tools, from the $49 Starter plan, cover your tracking grid across ChatGPT, Perplexity, Gemini, Claude, Grok, Google AI Overviews and Google AI Mode (get_visibility), the answers behind each cell (get_engine_answers), the pages engines cite (list_citation_sources), audits, gaps, the content queue, articles, competitor discovery, and ten Google Search Console analyses such as gsc_striking_distance and gsc_zero_click_pages. The keyword dataset behind list_keywords needs Pro.

The 9 write tools need Pro at $119: run_scan, set_prompts, set_gap_status, queue_article, mark_article_published, update_article, enrich_article, skip_queued_topics and set_outreach_status. The hosted connector is https://mcp.cituna.com/mcp: you paste the URL into Claude or another client, sign in to Cituna and approve access, and unticking the write box makes the connection read-only even on Pro. For scripts and CI, npx -y cituna-mcp runs it locally with a CITUNA_API_KEY token and Node 20 or later. The same data is on a REST API at cituna.com/api.

Strength: the daily history and the Search Console join in one server, plus a fresh scan when you need one.

Limit: the 3-day trial has no MCP access, Starter is read-only, and Microsoft Copilot is not among its engines.

Pick it if: you want an assistant to find the gap, check it against Google clicks, and queue the fix.

2. Cite42 (best for pay-per-check answers with no subscription)

Cite42 runs new checks on any prompt across ChatGPT, Claude, Gemini, Perplexity and Google AI Overviews and charges per call (its comparison page, dated 5 October 2026, read 22 October). A rankings or citations check costs $0.05 on ChatGPT, $0.08 on ChatGPT and Perplexity, and $0.20 on all five. You top up a prepaid balance, minimum $25, that does not expire, and a verified sign-up includes $1 of free credit.

Its tools are cite42_search for full answers and sources, cite42_rankings for mention rate and first-mention position, cite42_citations for cited URLs, and cite42_compare and cite42_sentiment, plus tracker tools that save up to 30 prompts and rerun them weekly or monthly with stored history. The hosted server signs in with OAuth, and an npm package runs it locally with an API key.

Strength: a fresh answer from five engines for 20 cents, with no plan to buy.

Limit: no database of past answers beyond your own trackers, trackers no more often than weekly, and no Copilot, AI Mode or Grok.

Pick it if: you want to check a prompt or a fix from inside the assistant, occasionally, without a subscription.

3. Manifold (best for free on-demand checks)

Manifold's hosted server includes three AI answer tools inside a wider server for SEO, social, Reddit, ads and leads: aeo_run_ai_answers runs prompts across engines as an uncached task, aeo_search_prompts finds prompts by keyword or topic, and aeo_get_site_readiness checks whether AI engines can read a site (its comparison page, updated 9 October 2026, read 22 October). It covers ChatGPT, Claude, Gemini, Perplexity, Google AI Overviews and Google AI Mode, and not Copilot. You sign in once, and calls spend credits: 250 free a month, with Pro from $20 a month.

Strength: a free way to see what six engines say now, from the same server as other marketing data.

Limit: no scheduled tracking or trend history; your assistant keeps whatever history you want.

Pick it if: you want spot checks at no cost and already use, or want, a general marketing MCP server.

4. LLM Pulse (best for the most tools on every plan)

LLM Pulse's hosted server, at https://api.llmpulse.ai/api/v1/mcp, lists about 130 tools across visibility metrics, share of voice, mentions, citations, sentiment, prompts and responses (read 22 October 2026). It signs in with OAuth 2.1 by default, and API keys for headless use need the Scale plan.

Write tools need the mcp:write scope: create_prompts, create_competitor, create_collection, assign_prompt_tags, create_annotation, create_intelligence_task, launch_recommendations and create_technical_geo_report. The two costly ones, launch_recommendations and create_technical_geo_report, return a preview and wait for your confirmation inside LLM Pulse. MCP is on every plan, the 14-day trial included, though AI traffic data needs Growth and agent analytics Scale. Starter costs €49 a month checked weekly or €79 daily, before VAT, for 50 prompts on five engines (read 13 October 2026).

Strength: the widest tool set here on the cheapest subscription, with a trial that includes MCP.

Limit: 130 tool descriptions take up room in an assistant's context, and Claude and Grok are add-ons.

Pick it if: you want to try MCP during a trial and need deep reporting from the first plan.

5. GEOly (best for write access granted per resource)

GEOly documents a hosted server at https://app.geoly.ai/api/mcp/v1 with up to 53 tools and OAuth sign-in (its server documentation, read 22 October 2026). Seven write tools, which create prompts, topics and competitors, archive or restore prompts, edit prompt tags, move prompts into a topic and trigger monitoring, are granted per resource at the consent screen; read is preselected and write stays off unless you tick it. Static tokens and connections that span several workspaces are always read-only. Market-intelligence tools need the Grow plan or above. Its engines are ChatGPT, Perplexity, Google AI Mode, Google AI Overviews, Gemini and Copilot.

The plan gate matters more than the tool count. GEOly's Basic plan, $49 a month for 30 prompts on ChatGPT only, includes MCP for custom brand data only, with a 3-day trial and no card. Grow, at $149, covers 50 prompts on ChatGPT and AI Mode with 500 tool calls a month; Advanced, $399, has 100 prompts on four engines and 1,000 calls; and Plus, $999, has 500 prompts on six engines, 5,000 calls and a REST API (read 22 October 2026).

Strength: the clearest consent model here: you choose which resources the assistant may change.

Limit: one engine at $49, and tool calls are metered by plan.

Pick it if: you want fine-grained write permissions and ChatGPT and AI Mode are your engines.

6. Peec AI (best for MCP on any paid plan, team-wide)

Peec AI's server is at https://api.peec.ai/mcp, and you connect by signing in with a Peec AI account that has a project set up (read 22 October 2026). Its page does not state the tool count, which tools write, or plan limits. Cite42's comparison, dated 5 October 2026, reports OAuth or a personal access token and write tools that work in batches of up to 50; we did not confirm those on Peec's own page.

Peec AI's pricing includes MCP on any paid plan, from Starter at $95 a month, or $80 billed annually, for 50 prompts checked daily on three models you pick, with unlimited users (read 1 to 9 October 2026).

Strength: MCP for a whole team on a flat plan with sentiment and share of voice.

Limit: little published detail about the server itself, and Claude and Grok only on Enterprise.

Pick it if: your team already reports in Peec and wants the same data in Claude.

7. Geoptie (best for half read, half write tools)

Geoptie's hosted server, at https://mcp.geoptie.com/mcp, has 48 tools: 24 that read and 24 that change data (read 22 October 2026). It signs in with an API key sent as a bearer token, with no OAuth, and allows 120 requests a minute per key. MCP is on Professional at $99 a month and Enterprise at $399; the Starter plan has no API or MCP. The MCP page does not state a trial. On 28 September 2026, Geoptie's Starter tracked four engines you pick from seven.

Strength: the largest published write set relative to its size, at $99.

Limit: an API key to manage rather than a browser sign-in, and no MCP below $99.

Pick it if: you want an assistant to change as much as it reads, and you are comfortable storing a key.

8. Otterly.AI (best for adding prompts and audits from chat)

Otterly.AI's server is at https://data.otterly.ai/mcp over streamable HTTP, with OAuth 2.0 sign-in and no API key (its help article, read 22 October 2026). Write tools ask for your permission: they create and update brand reports and annotations, create and delete prompts and tags, and trigger GEO audits and query fan-out. The help article does not give a tool count, and LLM Pulse's comparison counts 17 tools, 6 of them writes. Limits vary by plan, and trial accounts get limited MCP access.

MCP starts on Standard at $189 a month, with 100 prompts and 2,000 MCP requests a month; the $29 Lite plan has none (read 7 October 2026).

Strength: a daily program you can extend from the assistant, prompts included.

Limit: MCP starts five times above the entry price, and the server deletes as well as creates, so review what you approve.

Pick it if: you already run Otterly.AI on Standard and want Claude to manage it.

9. Scrunch (best for editing prompts, personas and competitors by chat)

Scrunch's developer documentation describes a server you sign in to with your Scrunch login in the browser, with no API keys (read 22 October 2026). It reads metrics and responses, including a list_responses tool, and writes brands, competitors, personas and prompts, imports agent traffic logs and handles migration imports. Each request returns up to 1,000 records, and cited URLs come only through raw responses.

The documentation lists any plan as the prerequisite, while Scrunch's pricing page, read on 1 October 2026, lists MCP access as custom on Enterprise. Core costs $250 a month for 125 prompts on four engines with a 7-day trial. Ask Scrunch which applies before you buy.

Strength: persona and competitor management from chat, which few servers here document.

Limit: conflicting plan terms, and no tool count published.

Pick it if: you are a Scrunch customer and want to manage your setup from an assistant.

10. Profound (best for enterprise teams already on Profound)

Profound's hosted server is at https://mcp.tryprofound.com/mcp over streamable HTTP. Clients sign in with OAuth when prompted, or use a long-lived bearer API key (read 22 October 2026). The page does not state the tools, which ones write, or the plan that includes it. Profound's pricing page lists the API on Enterprise, a custom contract with nine engines, and its free 7-day trial has no API (read 1 October 2026).

Strength: assistant access to Profound's enterprise data, with OAuth or a key.

Limit: no published tool list or plan gate, and no self-serve way in.

Pick it if: you already have a Profound Enterprise contract. Cheaper routes are in Profound alternatives.

MCP servers inside SEO suites: SE Ranking, Ahrefs and Semrush

If your team already pays for an SEO suite, its MCP server may give your assistant AI visibility data without a new tool.

SE Ranking hosts a server at https://api.seranking.com/mcp with OAuth 2.1 and dynamic client registration, or an X-Api-Key header (read 22 October 2026). It covers the SE Visible API, meaning AI visibility projects, brands, topics, prompts and sources, plus SE Ranking's Data API. The plan depends on a subscription with API access; SE Ranking's Core plan, $129 a month, includes MCP and 25,000 API credits (read 1 October 2026).

Ahrefs hosts a server at https://api.ahrefs.com/mcp/mcp with 20 tool groups that you select with a tools query parameter (read 22 October 2026). Two cover Brand Radar: brand-radar-simple for domain and URL reports and brand-radar-entities for entity reports. A management group covers projects, keyword lists, locations and Brand Radar configuration; the page does not say which tools change data. Picking groups keeps the tool list short. Ahrefs plans from Lite at $129 include API and MCP access (read 7 October 2026).

Semrush hosts a server at https://mcp.semrush.com/v2/mcp with OAuth by default or an API key, on Semrush One plans, SEO Classic Pro and Guru, Business or Advanced with an API units package, and Trends API subscriptions (read 22 October 2026). It exposes the Trends API, the SEO API and a read-only Projects API. The developer page does not mention the AI Visibility Toolkit, so on our read, Semrush's MCP server is not a route to its AI visibility data.

Other AI visibility tools that name an MCP server

These vendors name MCP on a pricing page or document a smaller server, without the detail the numbered entries have.

• Promptwatch includes MCP and API on its $95 Essential plan (read 8 October 2026); its MCP page returned an error on 22 October 2026, so sign-in and tool count are unconfirmed.
• Rankscale includes MCP on every plan, from Pro at $99 a month or Essentials at $204 a year; the REST API starts on Pro+ at $229 (read 17 October 2026).
• Trakkr includes MCP on its $100 Growth plan, for 50 prompts on eight engines daily (read 17 October 2026).
• Omnia includes MCP with read and write access on Growth at €79 a month, with a 14-day trial and no card (read 1 October 2026).
• Rankshift includes MCP and API on every plan from $92 a month (read 8 October 2026).
• Writesonic ships a read-only MCP server on every paid plan, from $99 (read 28 September 2026).
• Goodie AI includes MCP with 50 optimization actions a month on Core, $354 billed monthly (read 17 October 2026).
• Asva AI documents a hosted server with 17 tools, 15 that read and 2 that generate (generate_visibility_audit and generate_llms_txt), OAuth 2.1 with PKCE, and engines including Copilot, Grok and Meta AI, but no plan or price (read 22 October 2026).
• maxaeo publishes a free, MIT-licensed local server, npx -y maxaeo-ai-visibility-mcp, with three read-only tools that check a site's readiness, such as check_llms_txt; it does not query AI engines (read 22 October 2026).
• AthenaHQ does not mention an MCP server in its published plans, and sells API access as an add-on (read 1 October 2026).

Which AI visibility MCP server is best for agencies?

Agencies need one connection across many clients and a clear rule on what the assistant may change. GEOly's multi-workspace connections are read-only by design, which is safe for reporting. Cituna's Max plan, $399 a month, puts 10 brands on seven engines behind one connection, with write tools for queueing each client's fixes. More are in the best AI visibility tools for agencies.

Which AI visibility MCP server is best on a small budget?

For nothing, Manifold's 250 free credits a month and maxaeo's local readiness checker. For a few dollars, Cite42's per-check pricing, where a five-engine check is 20 cents. For a daily program with MCP, Cituna's $49 Starter is the lowest here with seven engines and Search Console, and LLM Pulse's €49 weekly Starter includes MCP from the trial. See the cheapest AI visibility tools for tracking without MCP.

Which AI visibility MCP server is best for enterprise teams?

Profound and Scrunch sell MCP inside enterprise contracts, and SE Ranking, Ahrefs and Semrush host servers inside suites procurement teams often already approve. Ask each vendor three things the public pages do not answer: request limits, how long history is kept, and whether write actions are logged.

Which AI visibility MCP server is best for a SaaS team?

A SaaS team often wants AI visibility where engineers already work: Claude Code, Cursor or a CI job. Cituna's local npx -y cituna-mcp and REST API suit that, beside the hosted connector for everyone else, with Search Console data for the pages that earn clicks. If your buyers ask Copilot, add GEOly, Otterly.AI or Peec AI, because Cituna does not track it.

How do you test an AI visibility MCP server before you rely on it?

1. Check the gate first. Confirm the plan includes MCP and whether the trial does: Cituna's and Peec AI's trials do not, LLM Pulse's does, and Otterly.AI's trial is limited.
2. Connect read-only. Untick or decline write access on the consent screen, where offered, until you have seen what the tools return.
3. Ask one question you can verify, such as "Which pages did Perplexity cite for our top prompt last week?", and check it against the vendor's dashboard.
4. Ask a trend question, such as "Did our ChatGPT mention rate change this month?" A stored-data server answers from history; an on-demand server needs a tracker or several calls.
5. Count the cost. Note the requests, credits or per-call charges one session used, against the plan's monthly allowance.
6. Turn on writes last, one tool at a time. Start with the least risky write, such as adding a prompt, and review it in the dashboard before you allow deletes or paid actions.

To size the prompt list you test with, see how to choose prompts to track.

Frequently asked questions

Which AI visibility tools have an MCP server?

Of the tools on this page, Cituna, LLM Pulse, GEOly, Geoptie, Peec AI, Otterly.AI, Scrunch, Profound, Cite42, Manifold, Asva AI and maxaeo document a server, and Promptwatch, Rankscale, Trakkr, Omnia, Rankshift, Writesonic and Goodie AI name MCP on their plans. SE Ranking, Ahrefs and Semrush host servers inside their SEO suites (read 1 to 22 October 2026).

Which AI visibility MCP servers can write, not only read?

Cituna (9 write tools on Pro), LLM Pulse (8 behind a write scope), GEOly (7, granted per resource), Geoptie (24), Otterly.AI, Scrunch and Asva AI (2 that generate) document write or change tools (read 22 October 2026). Writesonic's server and maxaeo's are read-only.

Is MCP included in AI visibility free trials?

Sometimes. LLM Pulse's 14-day trial includes MCP, Otterly.AI gives trial accounts limited access, Manifold and Cite42 have free credit, and GEOly's Basic plan has a 3-day trial. Cituna's 3-day trial does not include MCP, which starts with the $49 Starter plan (read 22 October 2026).

Does Semrush's MCP server include AI Visibility Toolkit data?

Not on our read. Semrush's developer page, read on 22 October 2026, lists the Trends API, the SEO API and a read-only Projects API, and does not mention the AI Visibility Toolkit.

Can Claude run a new AI visibility check through MCP?

Yes, with an on-demand server or a write tool. Cite42 and Manifold run new checks per call, and Cituna's run_scan starts a fresh scan of a tracked brand on Pro, using one scan from the plan's quota (read 22 October 2026). Stored-data servers return what their schedule already collected.