Bring your agent.
Your AI visibility data, in the assistant you already use.
The Prominara MCP server is a remote Model Context Protocol server that connects your own AI assistant to your Prominara workspace, a GEO platform that tracks brand visibility across AI answers. It is included in every Prominara plan and in the 14-day trial, so any customer can connect Claude, ChatGPT, Cursor or Codex today.
Ask what you are cited for, run a check, log the page you shipped, read the verdict. Same data, same quotas, no new dashboard to learn.
OAuth sign-inRevoke any timeIncluded in every plan
Add it as a remote MCP server in your client, sign in to Prominara, and choose read or read and write. Nothing else to configure.
- 01Add the server to your client
- 02Sign in and choose what the agent may do
- 03Ask your agent about your AI visibility
Founder, ProminaraUpdated
What can my agent do?
14 tools, split into reading your data and acting on it. Write is a separate permission you can leave off.
Reads your workspace
Available with the read permission. None of these spend quota.
List sites
prominara_list_sitesLists the sites (brands) in the connected Prominara organization with domain, brand name, latest audit score, last scan date and active prompt count. Call this first to learn which site identifiers to use with the other tools; sites can be addressed by domain or id.
Visibility overview
prominara_visibility_overviewReturns AI-search visibility for one site over a window (default 30 days): citation and mention rates with confidence intervals, per-platform breakdown, trend versus the previous window, and the competitor ranking. Rates use earned prompts and successful checks only. Use concise for headline numbers, detailed for per-prompt rows.
List tracked prompts
prominara_list_promptsLists the prompts tracked for a site with category, panel (organic, branded, competitive), active state and the latest result per platform. Filter by status or search text. Use it to find prompt ids before calling get_answers, run_check or set_prompt_active.
Get latest AI answers
prominara_get_answersReturns the latest recorded AI answers for one prompt: per platform whether the brand was mentioned or cited, the sources the answer used, position and sentiment, plus check history. Detailed mode includes the answer text, returned as untrusted third-party content.
List missions
prominara_list_missionsLists the prioritized missions for a site: for each tracked prompt, the current visibility, the page that should own the answer (or the gap), the source gap and the next best action, ranked by priority. Filter by priority, status or intent.
Sources intelligence
prominara_sourcesReturns the off-site sources behind AI answers for a site. Views: domains (which sites AI answers cite and whether they mention the brand), citations (individual cited pages), opportunities (ranked places to earn a citation) and accuracy (factual issues found in answers about the brand).
Measurement verdicts
prominara_measurementReturns the impact measurement for content changes logged on a site: per change, the observation window status and the verdict (likely lift, no effect, still collecting) with its probability, plus the program-level before/after comparison when enough data exists. Pass a change id for the detail of one change. Verdict wording is the product's own and does not claim causation beyond what the design supports.
Suggest prompts to track
prominara_suggest_promptsReturns suggested prompts for a site from the coverage matrix (organic, branded and competitive panels across buyer-journey stages), with a recommended set for the plan's slots. It may reuse cached suggestions or compute a fresh response, but does not save suggestions or tracked prompts; use prominara_create_prompts to keep selected prompts.
Plan and usage
prominara_usageReturns the organization's plan, monthly check quota used and remaining with the reset date, tracked prompt slots used, website and client-workspace allowances, platform slots and trial state. Call it before running checks or creating prompts to know what fits.
Acts on your behalf
Needs the write permission. Each one reports what changed and what quota is left.
Create tracked prompts
prominara_create_promptsAdds up to 20 prompts to track for a site and triggers their first checks. Duplicates of existing prompts are skipped, prompts beyond the plan's slots are skipped and reported, and platforms are clamped to the plan. Returns what was created, what was skipped and why.
Run a check now
prominara_run_checkRuns a fresh visibility check for up to 10 prompts across the site's enabled AI platforms, exactly like "Run now" in the app. Each check counts against the monthly quota. Prompts with a check already in progress are skipped and reported. Results arrive asynchronously; read them with get_answers after a minute or two.
Log a shipped content change
prominara_log_content_changeRecords that a page on the site was changed (edited, published, structured data added) with the release date and a summary, so Prominara can measure its impact on AI visibility. Creates the before/after experiment and freezes the affected prompts. Logging the same page and date twice returns the existing change.
Update mission status
prominara_update_missionSets the status of one mission (per prompt) on a site: new, in_progress, shipped, snoozed (with a date) or dismissed. Use shipped after the page work is live, or log_content_change which does it automatically.
Enable or disable prompts
prominara_set_prompt_activeActivates or deactivates up to 50 tracked prompts. Deactivated prompts stop being checked and free a slot; activating re-checks the plan's prompt slots and reports any that do not fit.
What does that look like in a conversation?
Five things people ask their assistant, and what it reaches for.
“Which of my tracked prompts lost citations this month, and which page should I fix first?”
Reads the visibility overview for the window, pulls the ranked missions, and answers with the prompts that moved and the page that should own each answer.
“Add five prompts about serverless database pricing and check them now.”
Creates the prompts against your plan slots, skipping duplicates, then runs the first checks. It tells you what was created, what was skipped and how much quota is left.
“I published the new comparison page today. Log it against the prompts it targets.”
Records the content change with its release date, freezes the affected prompts and opens the before-and-after observation window.
“Did the page I shipped three weeks ago move anything yet?”
Returns the measurement verdict for that change with its probability, range and evidence grade, in Prominara’s own wording.
“Which domains does ChatGPT cite in my category, and do any of them mention us?”
Reads the sources view for the site and lists the cited domains, whether each one names your brand, and the ranked places to earn a citation.
How do I connect?
Pick your client. Every route ends at the same consent page, where you choose the organization and the permissions.
Claude (web and desktop)
Add Prominara as a custom connector, then sign in once.
- 01Open Settings, then Connectors.
- 02Choose Add custom connector.
- 03Paste https://mcp.prominara.com as the server URL and save.
- 04Claude opens a Prominara consent page. Pick the organization, choose read or read and write, and approve.
How is it secured?
The connection is yours to grant and yours to end. Everything an agent can do, you could already do in the app.
OAuth 2.1, not pasted keys
You sign in to Prominara in your browser with PKCE. There is no key to copy into a config file, and nothing long-lived sits on your machine.
Consent on every connection
Each new connection shows the client asking, where it will send you back, and the organization you are connecting. Nothing is approved silently from a previous grant.
Read and write are separate
Write is a scope you can leave unchecked at consent. A read-only connection can look at everything and change nothing.
No tool deletes anything
The catalog covers reads and safe writes. Nothing touches billing, team settings, deletion, or another organization.
The same quotas as the app
A check your agent runs counts exactly like pressing Run now, against the same monthly allowance. Refusals say which limit was hit and when it resets.
Revoke whenever you want
Every connection is listed under Settings, Connected agents, with the client, the scopes and when it was last used. Revoking takes effect on the next call.
Choosing a GEO tool for your agent.
Generative engine optimization tools are starting to ship agent access. Five questions worth asking before you wire one into your assistant.
- 01
Sign-in, not key-pasting
Prefer OAuth. A tool that asks you to paste a long-lived key into a config file puts a credential on disk that nobody can revoke from the product.
- 02
Read and write, not read only
An agent that can only report leaves you doing the work by hand. Creating prompts, running checks and logging what you shipped is where the time goes.
- 03
Verdicts, not just counts
A citation count answers what happened. Whether the page you shipped moved anything needs a before-and-after design with a control group.
- 04
No separate usage units
Agent calls should draw on the plan you already pay for, not on a second meter with its own pricing.
- 05
Available on the cheapest plan
Check the entry tier, not the top one. Prominara includes agent access from Starter at $29 per month and in the 14-day trial.
What an agent can and cannot tell you.
Putting a language model in front of measurement data invites overclaiming. Here is where the line sits. Read the methodology
The verdict is the product’s, not the model’s
Measurement wording comes from Prominara and ships as a probability with a range and an evidence grade. Your agent relays it. It does not compute a verdict of its own from the numbers it reads.
A citation rate is a measurement, not a ranking
Rates describe the prompts you track, on the checks that succeeded, using questions that never name your brand. There is no placement to buy and none is promised.
Answer text is third-party content
When a tool returns what an AI engine said about you, it is marked as untrusted third-party text so your agent treats it as evidence to read, never as instructions to follow.
No timeline is implied
Checks run asynchronously and results land when the platforms answer. Nothing here states how long a content change takes to show up, because that is not ours to promise.
Before you connect.
The docs go deeper on each client, the full tool reference, and what to check when a connection will not authenticate.
Connect your AI agent docs01What is the Prominara MCP server?
02Which plans include it?
03Which AI clients work with it?
04Do I need to write any code?
05Can my agent delete or break something?
06Does an agent check cost more than a check in the app?
07What does the agent actually receive?
08How do I disconnect an agent?
Ask your agent.
It already knows how to work. Give it the data.
Machine-readable version of this page: /mcp.md