TofuBofu inside your assistant
Run your AI visibility scan without leaving the chat
TofuBofu publishes a remote MCP server. Paste one URL into your assistant and it gains the ability to run a real AI visibility scan for any B2B company, inside the conversation: no signup form first, no API key, nothing to install. The window your buyers use to research vendors becomes the window where you check whether they can find you.
Every plan queries all 6. See how each engine picks who it names
What we measured
of B2B buyers now begin vendor research on an AI chatbot, up from 29% a year earlier (G2, 2026). The MCP server puts the visibility check inside that same window.
Engines queried
ChatGPT, Claude, Perplexity, Gemini, Google AI Mode, Microsoft Copilot. Every plan, including the free one.
Answers kept
Every answer is stored word for word, so any number on your report can be checked against the text it came from.
The gap this closes
- Checking AI visibility meant leaving your tool, finding a site, filling a form and waiting for mail.
- Every one of those steps loses people who were already curious.
- The question gets asked inside an assistant, mid-conversation, not on a marketing site.
- By the time a form is involved, the moment has passed.
What you end up with
- A real scan run from inside Claude, ChatGPT or any MCP client, free to connect.
- The answer arrives in the conversation where the question was asked.
- The same scan the website runs, not a demo or a trimmed preview.
- Setup measured in one config block.
How it works
Three steps, and you only do the first one.
Connect the URL once
In Claude, open Settings, then Connectors, then Add custom connector, name it TofuBofu and paste https://tofubofu.com/mcp. Custom connectors need a paid Claude plan. In Claude Code one command does it: claude mcp add --transport http tofubofu https://tofubofu.com/mcp. ChatGPT, Cursor and Windsurf take the same URL in their own connector settings.
Ask in plain language
Say something like: use TofuBofu to check AI visibility for acme.com, my email is you@company.com. The assistant calls scan_ai_visibility with those two values and tells you the scan takes a minute or two. Nothing else is requested and nothing else is needed.
Read the report back in the same conversation
Ask for the report and the assistant calls get_visibility_report, then answers follow-up questions from the returned data: which questions you lose, who gets named instead, which engine was worst. The live report link works in a browser and the finished report arrives by email.
What you get on each plan
Connecting is free and the scan it runs is the real one, not a demo. This is what the server gives you, per plan, with the limits read live from the plan configuration.
Track
Free- Both tools work for anyone with the URL
- No account, no key
- A domain and a work email runs a genuine six-engine scan
- Report read back in the chat a minute or two later
- One scan per company per 30 days, same as the website
- Emailed, so it survives the chat window closing
Fix
$99/mo- The MCP scan is the front door; history lives in the app
- Weekly scans, sentiment and drafted pages need an account
- CMS publishing and alerts need a record over time
- A chat-started report is already attached to your email
Dominate
$499/mo- Wider budget and competitor detail change what a chat scan is worth
- Scan a prospect's domain during the call
- Walk into the pitch holding their gaps
- Keep the ones who convert on a tracked cadence
Rule
$1999/mo- Portfolio level, a scan per brand rather than per conversation
- Chat stays the fastest way to check something on the spot
- The app is where the portfolio view lives
The proof, including the parts that flatter nobody
The scan the server runs is the scan behind every number we publish. Across the 46 completed reports in that corpus, the best any engine managed was naming the brand in 7.4% of the buying questions we put to it, and the others sat between 0.4% and 4.0%, on firms that arrived already suspecting they were missing.
The server is listed where MCP clients look: the official registry, mcp.so, and a Custom GPT in the GPT Store for people who work in ChatGPT. Those listings are linked below so you can verify them rather than take our word for it.
What the free scan does not do is track you over time. One scan a month is a presence check, each question runs once instead of being sampled, and sentiment stays off. That is a real limitation, it is the honest shape of a free tier, and it is stated here rather than discovered later.
Read the MSP AI Visibility Report 2026 →A worked example: the whole exchange
Below is the shape of a real conversation, with an illustrative domain. Nothing is elided except the wait, which runs a minute or two while six engines answer.
In the chat
You: Use TofuBofu to check AI visibility for northlake-it.com,
my email is dana@northlake-it.com
Claude: [calls scan_ai_visibility]
Started. This runs real queries across six engines and takes
about a minute. I will read the results when they are ready.
You: get the report
Claude: [calls get_visibility_report]
Northlake IT appears in 1 of the 5 buying questions I tested,
and only on Perplexity. The question you lose most clearly is
after-hours coverage for manufacturers: four engines answered
it and three named the same national provider. One of those
answers said to look for a documented shift-change handover,
which you do and do not publish anywhere.
You: which engine was worst, and what would you fix first?
Claude: [answers from the report data already returned]The same call over plain JSON-RPC, for anyone who wants to look
curl -s https://tofubofu.com/mcp -H 'content-type: application/json' -d '{"jsonrpc":"2.0","id":1,"method":"tools/list"}'What the server received in that exchange
- One domain, typed by the user.
- One email address, typed by the user.
- One report identifier, returned by the first call.
- Nothing else. No files, no account, no other connector, no browsing history.
The last block is the point of the whole design. A tool that could reach further would be harder to trust and no more useful, so it does not reach further.
How it actually works
The mechanism, for anyone who wants it. Open a card to read the detail.
What MCP is, and what a server is allowed to do
The Model Context Protocol is an open standard, introduced by Anthropic, that lets an assistant call external tools during a conversation. A remote server is a URL the client connects to; once connected, the assistant can use the tools that server declares, and only those.
Read the detail
That last clause is the security model and it is worth understanding before you connect anything. A server declares a fixed set of tools with typed inputs. It cannot reach into your files, your other connectors, your account or anything you have not typed into the conversation. It receives the arguments the assistant passes and nothing else.
Our two tools also carry the behavioural annotations the specification defines, so a client can tell what they do before calling them. Both are marked read-only and neither is destructive, which is a claim the client can display to you rather than one you have to take from a marketing page.
The tool contract, stated in full
scan_ai_visibility takes two arguments: the company domain you want scanned and a work email. It starts a real scan across ChatGPT, Claude, Perplexity, Gemini, Google AI Mode and Microsoft Copilot, and returns an identifier plus an estimate of how long to wait. The email is where the finished report is sent, and it is how the one-per-30-days fair-use limit is applied.
Read the detail
get_visibility_report takes the identifier and returns the finished result: the score, the per-question findings, the competitors named in the answers and a link to the live report. Call it too early and it tells you the scan is still running rather than returning a half-formed answer.
There is no third tool and no hidden surface. If you want to inspect that claim rather than accept it, point the MCP Inspector at the endpoint, or send JSON-RPC over plain curl. Opening the URL in a browser shows a setup page instead of the protocol stream, so the link is safe to send to a colleague.
Why this belongs in the chat and not behind a form
A vendor list generated by an assistant is the thing being measured, so running the measurement in the same window is not a convenience feature. It is the shortest path from noticing a problem to seeing evidence of it, and the evidence arrives in a format the assistant can then reason about with you.
Read the detail
The pattern that has surprised us most is agencies. Scanning a prospect's domain during a call, reading the gaps aloud, and leaving the live report link behind turns a pitch into a diagnosis. The fair-use limit of one scan per company per 30 days applies the same way, and the report works as a leave-behind because it needs no account to open.
The transport is streamable HTTP, the current standard, which is why the same URL works everywhere without a per-client build. That was a deliberate choice: a server that needs a different setup in every client is a server most people never connect.
What it costs
Connecting the server costs nothing. The plan below decides what the scan behind it includes, read live from the plan configuration.
| Plan | Price | Questions | Scans |
|---|---|---|---|
| Track | Free | 5 | 1/mo |
| Fix | $99/mo | 10 | 4/mo |
| Dominate | $499/mo | 25 | 4/mo |
Rule is portfolio level and sales-assisted, and it suits a group running scans across several brands.
Find TofuBofu across the AI ecosystem
See where you stand first
Run a free scan and find out which of these gaps you actually have.
Get your free auditFrequently asked questions
What is MCP?
An open standard, introduced by Anthropic, that lets an assistant call external tools during a conversation. A remote server is a URL the client connects to, and once connected the assistant can use the tools that server declares and only those.
Is the TofuBofu MCP server free?
Yes. Connecting costs nothing and the scan is the same free scan the website runs: one per company per 30 days, all six engines, report emailed with a live link. Paid plans add cadence, sentiment, drafted content and history, and none of that is needed to use the server.
Do I need an API key or an account first?
No. The server needs the domain you want scanned and a work email, both passed conversationally. The email is where the report is sent and how the fair-use limit is applied. Create an account later and the report is already attached to it.
Is it safe to connect?
The server declares two tools, both annotated read-only and neither destructive, and it receives only the arguments you type. It cannot reach your files, your account, your other connectors or anything else in the conversation you did not give it.
Which assistants can use it?
Any client supporting remote MCP servers over streamable HTTP, which today includes Claude on the web, desktop and in Claude Code, ChatGPT connectors, Cursor and Windsurf. The URL is the same everywhere.
Can I scan companies other than my own?
Yes, and agencies do exactly that: scan a prospect mid-conversation and walk into the pitch holding their gaps. The one-per-company-per-30-days limit applies, and the report opens without an account so it works as a leave-behind.
What data does it collect?
The domain you asked to scan and the email you gave for delivery. No tracking cookies in the MCP flow and nothing else requested. The report is the same one the website produces.
Sources and further reading
- Model Context Protocol specification: The open standard for connecting AI assistants to external tools and data sources.
- G2 2026 AI Search Insight Report: 51% of B2B software buyers begin research on an AI chatbot more often than Google; 69% chose a different vendor than planned based on AI guidance.
- Official MCP Registry: The public registry MCP clients use to discover remote servers like this one.
Related solutions