MCP
Ask your own data,
from your own client
LLMagnet is an MCP server. Connect it to Claude, ChatGPT or any MCP client, and ask about your visibility, your gaps, your sources and your traffic, with your own scans behind every answer.
Book a demoSee how it connects
30-day free trial · no credit card · MCP and API access on Pro and above
A tool that measures what AI says about you should be something AI can talk to.
Four things to know
before you connect it
A standard server, your own account, one website at a time.
01Your client
A standard MCP server
+ any MCP clientAny client that speaks the Model Context Protocol can use it: Claude, ChatGPT, your editor, your own agent.
02LLMagnet
Authorized with OAuthAllow Claude to connect?LLMagnet account · you@brand.com
AllowYou approve the client against your LLMagnet account. No API key pasted into a config file.
03You
Scoped to one website- hardly.com
- hardly.co.uk
A connected client reads the website you choose, with the same plan limits as the app. Not your whole workspace.
04Your client
The tools appear- get_prompt_gaps
- get_sources
- save_article_to_cms
Ask in plain language. The client picks the tool, and shows you which one it called.
Pro and aboveMCP and API access come with Pro, Agency and Enterprise. Starter doesn’t include them.
What it looks like
in the client
One question, one tool call, one answer grounded in your scans, not in the model’s memory of your brand.
Claude
Which of our tracked questions did we lose on ChatGPT in the last two weeks?
get_prompt_gapsLLMagnet
- website
- hardly.com
- engine
- chatgpt
- since
- 14 days
Four of your tracked prompts no longer name Hardly in ChatGPT’s latest answer:
- “best squat-proof leggings”Lululemon and Alo Yoga named instead
- “leggings that don’t roll down”Named on Perplexity, missing on ChatGPT and Gemini
- “are seamless leggings good for lifting”No brands named in the answer at all
- “best gym shorts with a phone pocket”Named here two weeks ago; Nike is named now
The second is the clearest gap: you’re named on another engine for the same question, which usually means the content exists but isn’t in the form ChatGPT is quoting.
What’s cited in the answers we’re missing from?
get_sourcesLLMagnet
- website
- hardly.com
- prompts
- 4 gaps
Three of the four cite garagegymreviews.com, where you don’t appear, and two cite a Reddit thread. Only one cites a brand-owned page, and it’s Lululemon’s.
- 01The tool call is visible. You see which tool ran, and which website it was scoped to.
- 02The answer comes from your scans. Not from whatever the model happens to remember about your brand.
- 03Ask follow-ups. Here the conversation is the interface, so ask the question the dashboard has no view for.
Read everything.
Write nothing that publishes.
The MCP server is how you ask, and it can save a draft of what you decide. It is never how anything goes live.
What a client can read
- Visibility scores by engine
- Tracked prompts, and the answers you’re missing from
- The competitors named instead of you
- The sources cited in answers about you
- Site audit and crawlability findings
- AI traffic and conversions
- Your reports
What it can write
Around fifty of its tools can write, across ten write scopes. Almost all of them change data inside LLMagnet: adding a competitor, adding a topic, creating a report.
save_article_to_cmsDraft only
WordPress, the Shopify blog or a Webflow collection
The one tool that writes outward. It saves a draft. Nothing is published.
What it will never do
Publish anything, anywhere.
Every write tool needs write access, so a viewer seat can’t call one. Anything beyond a saved draft is prepared by the agent and runs only after you approve a card inside LLMagnet, never from a chat in another client.
The numbers a client reads are the same consistent sample the app shows: your prompts, asked once per engine per scan, on your plan’s cadence. A client can state a number. It can’t make the sample bigger.
Read how we measureWhy ask from a chat, and not the dashboard?
Because that’s where a lot of the work already happens. If you’re drafting a brief or debugging a page in Claude, pulling your gap list into that thread beats opening another tab, and you can ask follow-ups the dashboard has no view for.
If you prefer the dashboard, use the dashboard. It’s the same data.
In LLMagnet41%ChatGPT visibility · 30 days
In Claude41%“Your ChatGPT visibility is 41% over 30 days.”
Same scans, same number
Ask from anywhere.
Fix it in LLMagnet.
MCP is how you find what’s wrong, from wherever you work. The agent is what does something about it.
Next · The agent
The gap you asked about, fixed
The agent prepares the change in your Shopify store, Webflow site or WordPress CMS, and waits for your approval. Nothing ships until you approve it.
Asked in Claude · “leggings that don’t roll down”
Say the waistband stays putRep Seamless Leggings · Shopify
A high, sculpting waistband that stays put through every squat and lunge.
- Named on Perplexity, missing on ChatGPT and Gemini
- The page never says the waistband doesn’t roll
Plans
MCP and API access start on Pro
The same plans as the app, with the same limits. No separate MCP plan.
- StarterNot included
- ProMCP & API
- AgencyMCP & API
- EnterpriseMCP & API
Action Center · where the gaps you ask about are rankedSite Audit · crawlability findings, readable from your client too
Frequently asked questions
Want to see the product before you start? Take the product tour, or start a 30-day free trial.
Any MCP client. We document Claude and ChatGPT because that’s what most people ask for; the server is standard, so your editor or your own agent works the same way.
Connect it to
your client.
OAuth, one website, your own scan data. Pro and above.
Book a demo
30-day free trial · no credit card