It reads
Eight read tools cover your brands, overview, engines, competitors, cited sources, prompts and run log.
AI visibility MCP for Claude and ChatGPT
Claude and ChatGPT read your AI visibility data. Ask about your share of AI answers, rank, competitors and cited sources in plain language, and get answers from your latest published run.
What is the AI visibility MCP server?
The short answer
The Firemark MCP server lets Claude, ChatGPT and other AI assistants read your results directly. It returns your latest published run, with answer snippets of up to 500 characters, and it never starts runs or touches billing.
It drafts
An assistant can interview you and draft a prompt set, or propose prompts and competitors to add or remove.
You confirm
A member of the account approves or discards every draft in the app. Nothing runs or is charged before that.
The numbers are the ones your screens show, counted by the rules on the method page. To see them first, open the demo report.
Connect it to Claude and ChatGPT
Add the server URL as a connector in your AI app, sign in with your Firemark account and approve access. The steps are the same idea in both apps, and neither needs an API key.
- Copy the server URLUse the address shown for your account, in the form below.
- Add a connectorIn Claude or ChatGPT, open the connector settings and add a custom remote MCP server with that URL.
- Sign inThe app opens the sign-in page. Use the account that holds your brands.
- Approve accessCheck the assistant's name and the permissions on the consent page, then choose Allow.
- Ask a questionStart with "Which brands do I track?" and go from there.
Server URL, placeholder
https://mcp.example.com [placeholder]What your assistant can read
Eight read tools return the latest published run, the same figures and sentences your screens show. Each result also tells the assistant how to read the numbers, including which engines were short.
| Tool | What it returns | Example question |
|---|---|---|
list_brands | The brands in the accounts you belong to, each with its state, last and next run, and how many prompts it tracks. Assistants call this first. | Which brands do I track, and when did each last run? |
get_overview | The headline for the latest run: answers naming the brand out of answers collected, the Overview sentence, outcomes, counts by funnel stage and by engine, and any movement or rank claim the rules allow. | How many AI answers named us in the latest run? |
get_engines | Each engine in the latest run: answers naming the brand, the share (on a base of 20 or more), the 95% range, counts by funnel stage, outcomes and the engine's history by run. | Which engine names us least? |
get_competitors | Each tracked competitor against your brand: answers naming it, by engine, head to head and rank. A competitor named in fewer than 10 answers has no figures. | Which competitor overtook us on Perplexity? |
get_sources | The sites the answers cite, with answers citing each one, split into answers that name your brand and answers that name a competitor and not you. Site level only. | Which sites does AI cite when it names a rival and not us? |
list_prompts | Your prompts with one verdict per engine for the latest run, filtered and sorted as the Prompts screen does. Snippets are off by default and limited to 10 rows when on. | Which prompts are we absent from on ChatGPT? |
get_prompt | One prompt with each engine's verdict, position, sentiment, competitors named, how often you were named over recent runs, and a snippet of up to 500 characters around your brand's name. | What does the AI say about us for this prompt? |
get_runs | The run log, newest first: date, number, status, coverage by engine, answers collected and engines left out, plus the next scheduled run. It works before the first run is published. | Did last night's run finish, and when is the next one? |
Tools that draft changes for you to confirm
Five more tools draft a prompt set or propose changes. They write a draft and a review link, and never change a brand.
| Tool | What it returns | Example question |
|---|---|---|
interview_start | Starts, or resumes, a five-step interview that ends in a draft prompt set, and says what to ask next. Nothing is run or charged. | Help me set up tracking for a new brand. |
interview_submit_step | Stores your answers to one step (brand, competitors, buyer types, offering, review) and returns the next. | Our competitors are these three sites. |
interview_get_basket | The prompts suggested so far, each with its funnel stage, buyer type and whether it names your brand or a competitor. | Show me the prompts you suggest. |
interview_save_draft | Saves the finished interview as a draft and returns a link where a member confirms it in the app. The draft tracks nothing until then. | Save this as a draft for me to confirm. |
propose_changes | For a brand that already runs: prompts and competitors to add or remove, written as a proposal with a review link. It never edits the brand. | Propose three prompts to add for our new product. |
What the AI visibility MCP server cannot do
An assistant connected through MCP cannot start a run, change your schedule, touch billing or members, or return a full answer. No tool on the list does any of these.
Start a run
Runs follow your schedule. No tool starts one.
Change schedule or engines
Cadence and engines are set by you and your plan in the app.
Touch billing or members
Plans, payments and who is on the account stay in the app.
Edit a brand or a confirmed prompt set
Assistants can only draft or propose. A member approves or discards in the app.
Return full answers
Only counts, verdicts and snippets of up to 500 characters. Cited pages are never returned.
Read other accounts
A brand you are not a member of answers exactly as one that does not exist.
Example questions to ask Claude or ChatGPT
Ask in plain language and the assistant picks the tool. These answers come from the demo report, so they match the numbers on the demo page.
For SEO and marketing leads
Demo data
SEO leadHow many AI answers named us in the latest run?
Assistant, via MCPNamed in 312 of 625 answers across five engines, a 50% share. You rank #2 of 6, behind Tool A, and your site was linked in 183 answers.Tool: get_overview
Demo data
SEO leadWhich engine names us least?
Assistant, via MCPGemini: 50 of 125 answers, 40%. ChatGPT is highest at 78 of 125, 62%, so the blended 50% hides a 22-point gap.Tool: get_engines
Demo data
SEO leadWhich competitor overtook us on Perplexity since the first run?
Assistant, via MCPTool B. It went from #4 to #1 on Perplexity, from 25 to 85 answers naming it. You went from #1 to #2, from 40 to 71.Tool: get_competitors
For founders and CMOs
Demo data
FounderIs AI sending buyers to us or to a rival?
Assistant, via MCPBoth. You are named in 50% of answers and rank #2, up from #4 in run 1. Tool A is still ahead, named in 361 of 625 answers, 58%.Tool: get_overview
Demo data
FounderWhere in the buying journey do we lose?
Assistant, via MCPResearch. You are named in 84 of 200 research answers, 42%, against 118 of 225 compare answers, 52%, and 110 of 200 decision answers, 55%.Tool: get_overview
Demo data
FounderWhich sites are cited when a competitor is named and we are not?
Assistant, via MCPTool A's own site in 98 answers, Review site C in 74 and Tool B's site in 57.Tool: get_sources
For ecommerce and retail brands
Demo data
Ecommerce leadWhich retailers and review sites does AI cite for our category?
Assistant, via MCPRetailer B is cited in 238 of 625 answers and Review site C in 226. Your own site is cited in 150.Tool: get_sources
Demo data
Ecommerce leadWhich product questions are we absent from on ChatGPT?
Assistant, via MCPMostly research and decision questions, such as whether carbon plate shoes make you faster, and whether Brand B is worth the price. Both name Brand B and leave you out.Tool: list_prompts
Demo data
Ecommerce leadDraft a prompt set for our new trail line.
Assistant, via MCPI asked five short questions about your brand, competitors, buyer types and offering, then saved a draft. Open the link to review it. Nothing runs or costs anything until a member confirms it.Tool: interview_save_draft
Permissions and security
You sign in with your own account and approve each assistant on a consent page, and the assistant can only read or draft within the accounts you belong to.
Sign in with your account
The server is its own OAuth 2.1 authorization server with PKCE. The assistant never sees your password, and there are no API keys.
A consent page you approve
After sign-in, a consent page names the assistant and what it would be able to do, with Allow and Do not allow.
Two permissions
Read your results, and draft a prompt set or propose changes. Draft changes take effect only when a member confirms them in the app.
Scoped to your accounts
Every call is checked against your membership. Leaving an account cuts access to it at once, with the same token.
Short-lived tokens
Access tokens last one hour. A refresh token lasts 30 days and rotates. A screen for revoking one connection is not built yet.
Treated as data
Quoted answer text comes back marked as data, not instructions, so an answer cannot give the assistant orders.
AI visibility MCP FAQ
MCP
How does MCP work, and which AI apps does it connect to?
MCP lets Claude and ChatGPT read your AI visibility results. Connect the Firemark MCP server to Claude or ChatGPT, sign in with your Firemark account, approve access, then ask questions in plain language. Read tools cover brands, overview, engines, competitors, cited sources, prompts and the run log.
What can an AI assistant do with my data through MCP?
It reads your latest published run, with answer snippets of up to 500 characters and no full answers. It can draft a prompt set or propose changes to your prompts and competitors, and you confirm them in the app. It cannot start runs, change your schedule or engines, or touch billing or members.
Is it safe to connect an AI assistant to my account?
You sign in with your own account and approve the connection on a consent page that names the assistant and what it can do. There are two permissions, read your results and draft changes, and nothing an assistant drafts takes effect until a member confirms it in the app. Removing a member from the account ends that person's access at once.
Can the assistant see the full AI answers or the pages they cite?
No. It gets counts, verdicts and a short snippet of up to 500 characters around your brand's name. Full answer text and cited pages are not returned, and cited sources are shown at site level only.
Does it work with AI apps other than Claude and ChatGPT?
The server is a standard remote MCP server that uses OAuth sign-in, so other assistants that support remote MCP servers can connect in the same way. Claude and ChatGPT are the two this page describes.
Why does my assistant say there is no data yet?
Read tools use the latest published run, so a brand has nothing to read until its first run finishes. Until then the assistant can still read the run log and see when the next run is scheduled.
Ready to connect? See the plans and start tracking, then add the server to your AI app.
Ask your AI app about your AI visibility
Start tracking, then connect Claude or ChatGPT with MCP.