Twelve of the sixteen AI visibility tools we checked on 17 August 2026 document an MCP server, so "does it connect to Claude Code" no longer separates them. The question that still separates them is what the connection carries: visibility data out to your agent, or a change back into your site.
Disclosure: GetIntel is an AI visibility tool and ships an MCP server, so we have a commercial interest here. Every vendor claim below is quoted from that vendor's own public documentation, checked on 17 August 2026. The method and its limits are set out at the end.
We started this as a shortlist piece, expecting a handful of vendors to support Model Context Protocol and the rest not to. The census killed that premise in about twenty minutes. MCP is now table stakes in this category, including for us, and publishing the list we planned would have told you almost nothing.
So this is the finding instead.
What is an MCP server, briefly?
Model Context Protocol is an open standard for letting an AI agent call an external tool. A vendor runs a small server, the server declares a list of tools it offers, and an agent such as Claude Code or Cursor can call them during a conversation. You configure it once, and from then on the agent can reach that vendor's data without you pasting anything into a chat window.
The practical effect for AI visibility work is that the measurement stops living behind a login. Instead of opening a dashboard, exporting a CSV and describing the problem to your agent, the agent queries the tool directly and works from the live numbers.
That is the part every vendor has now built. The part that varies is whether the same connection can send anything back.
Which tools document an MCP server?
Twelve of sixteen, checked 17 August 2026.
| Vendor | MCP documented |
|---|---|
| Otterly | Yes |
| Profound | Yes |
| Peec | Yes |
| Semrush | Yes |
| Ahrefs | Yes |
| SE Ranking | Yes |
| Conductor | Yes |
| Frase | Yes |
| LLM Pulse | Yes |
| Dageno | Yes |
| Kime | Yes |
| Ayzeo | Yes |
| Scrunch | Not found |
| AthenaHQ | Not found |
| LLMrefs | Not found |
| ZipTie | Not found |
The sample is the sixteen vendors that AI engines actually name in our own category tracking, not a directory listing. Several announced recently and loudly. Peec states "Peec MCP is live." Dageno states "Dageno MCP is Live. Connect Dageno to Claude, Cursor." Kime states "KIME MCP is now live." SE Ranking ships what it calls MCP SEO Skills for Claude.
Absence of documentation is not proof of absence. A vendor may ship an MCP server without publishing docs for it, so read the four negatives as "not documented on 17 August 2026" rather than "does not exist".
Why has MCP stopped being a differentiator?
Because the protocol is cheap to adopt and the data was already sitting there. Every tool in this category runs prompts against AI engines and stores the results. Exposing that store over MCP is a thin layer on an existing API, which is why a dozen vendors shipped one inside roughly the same window.
Twelve months ago the picture was different. Connecting a visibility tool to a coding agent meant writing your own glue against a REST API, and the vendors who removed that work first had something real to sell. Adoption caught up.
So you can stop treating MCP as a shortlisting criterion. It tells you very little about which tool fits your workflow, because nearly all of them clear the bar.
What does the connection actually carry?
Two things, and they are not the same thing. One kind of MCP sends visibility data out to your agent. The other carries a change back into your site.
Data out is the common case. Your agent asks the tool what it measured, and the tool answers. Peec's own documentation, checked 17 August 2026, describes exactly this: its MCP lets you "feed visibility data directly into AI agents such as Claude". That is genuinely useful. You can ask Claude to summarise last week's movement, compare two competitors, or assemble a report without opening a dashboard.
A change in is the rarer case. The agent does not only read the finding, it carries a fix back: an llms.txt file, schema markup, a counter-article. Frase's own site, checked 17 August 2026, offers to "let Frase publish on its own, keep every page under review, or anywhere in between". Ayzeo lists a WordPress MCP among its integrations, checked 17 August 2026, routing changes through that platform.
Both get described with the same three letters and solve different problems.
Where does the fix actually land?
In one of three places, and the destination matters more than the protocol.
- Into the vendor's own plugin. The change ships through software the vendor controls, which means your site has to run the platform that plugin targets.
- Directly onto your live pages. Fast, and it skips the review step your team already uses for every other change.
- Into your repository as a pull request. The change arrives in your normal review flow, a human approves it, and your existing deploy pipeline ships it.
Those are three different risk profiles, and none of them is visible from the phrase "MCP support".
GetIntel takes the third route. Our own feature page states that the drafted fix is handed "to Claude Code or Cursor via MCP" and "lands as a reviewable PR in your real repo". As of 17 August 2026 our live MCP server exposes twenty read tools alongside five act tools, so an agent can both read the gap and act on it with your sign-off.
How do you check this yourself?
Read the tool list, not the marketing page. An MCP server declares the tools it exposes the moment your agent connects, and that declaration is the honest answer to what it can do.
Connect the server in Claude Code or Cursor, then sort what appears by verb:
- Reads: names beginning with
get,list,fetchorsearch. A server offering only these is a reporting surface, which may be exactly what you want. - Acts: names beginning with
create,update,publish,approveorship. These can change something, and you should know what they can reach before you approve them.
Check what the connection costs before you wire it up. An MCP server needs credentials, and those credentials usually carry the plan limits of the account behind them, so a read that is free in the dashboard can consume metered API calls when an agent makes it fifty times in a loop. Confirm which plan tier exposes MCP at all, whether calls are metered separately from seats, and what the rate limit is before you let an agent run unattended against it.
Then check the destination, because the verb alone does not tell you where the change lands. Ask three questions of any vendor: where does a change go, do you see it before it goes live, and what happens if you reject it. A vendor whose documentation cannot answer those three is describing an intention rather than a workflow.
Worth saying plainly: we have not run that connect-and-inspect test across these servers yet. Connecting is also only half the problem, since a fix still has to be shipped by someone, which is the constraint we measured on our own brand in the action gap in AI visibility tools. This piece is a documentation census, so nothing here should be read as a claim about what any competitor's MCP can or cannot do once connected. That test is the follow-up, and we will publish the tool surfaces we find whichever way they fall.
How did we check this?
We fetched the public documentation and product pages of sixteen vendors on 17 August 2026. The sixteen were selected because AI engines name them in our own category tracking, not because they appear in a directory or a funded-startup list.
A vendor counts as documented when its own site either publishes MCP documentation or lists MCP as a shipped integration in its navigation or integrations page. Marketing copy promising MCP "soon" would not have counted, and we did not encounter any. We did not connect to any server or enumerate its tools, so the table above records what each vendor publishes about itself and nothing more.
The obvious limitation is that a vendor can ship a feature without documenting it. A second is that documentation goes stale, in both directions, which is why every figure here carries its date.
If you want to see the gap before you wire anything up, our guide to shipping AI visibility fixes without an engineering backlog walks the same loop end to end, and the B2B SaaS tool comparison covers how these vendors differ outside the MCP question.