New · Model Context Protocol

Ask ChatGPT or Claude about your AI visibility — and get the real numbers back.

The Aparok MCP server puts 15 tools inside the assistant you already use. Prompt performance, competitor citations, AI traffic, site audits, and prospect qualification — queried in plain language, answered from your verified workspace data.

Works as a ChatGPT app connector (formerly plugins) and as a Claude connector. Paste one URL, approve access, and the dashboard becomes conversational.

Start 7-day trial See setup guide

OAuth 2.1 + PKCE11 read-only toolsIncluded in every plan

In your assistantAparok connected

“Which buyer prompts are we losing this month, and who’s winning them?”

list_workspaces1 workspace

get_prompt_performance25 prompts

get_competitor_visibility12 competitors

9 of 25 tracked prompts are losing.

Citation rate fell 4.2pp week over week. Two review sites take the top slot on every “best… for” prompt — with the answer excerpts to prove it.

Why connect it

Your data should answer questions where you already ask them.

Dashboards make you go looking. A connector brings the answer to the conversation you are already having — and lets the assistant do the analysis, drafting, and follow-up for you.

No context switching

Ask in the chat you’re already in

Pull visibility, citations, traffic, and audit findings mid-conversation instead of opening a tab, picking a date range, and exporting a CSV.

Grounded, not guessed

Every answer carries its evidence

Tools return observation counts, comparison periods, answer excerpts, and source URLs, so the assistant reports what was measured rather than improvising a plausible number.

Analysis included

Go from data to deliverable in one turn

The same conversation that fetches your losing prompts can draft the content brief, the sprint plan, or the exec update — because the assistant already has the numbers.

Setup

One URL. About two minutes.

The server supports OAuth 2.1 with PKCE and dynamic client registration, so most clients ask for nothing but the address below.

Server URLhttps://aparok.com/api/mcp

TransportStreamable HTTP

AuthOAuth 2.1 (PKCE) or Bearer key

Scopesmcp:read · mcp:write · offline_access

Tools15 (11 read · 4 write)

ChatGPT

Apps & Connectors

Formerly known as plugins. Available on Plus, Pro, Business, and Enterprise; custom connectors may need to be enabled in developer mode or by a workspace admin.

Open Settings → Connectors.
Choose Create or Add custom connector.
Paste the server URL and name it Aparok.
Leave authentication set to OAuth — discovery is automatic.
Click connect, sign in to Aparok, and approve the scopes.
In a new chat, enable Aparok from the connector menu and ask it to list your workspaces.

Claude

Custom connector

Works on Claude web, desktop, and mobile. On Team and Enterprise plans an owner or admin can add Aparok once for the whole organization.

Open Settings → Connectors.
Click Add custom connector.
Paste the server URL and save.
Click Connect and complete the Aparok sign-in.
Approve mcp:read, and mcp:write if you want the assistant to queue runs.
Start a chat and ask which Aparok workspaces it can see.

Developers

Claude Code, Cursor & custom clients

For headless clients, mint an MCP API key and send it as a bearer token. Keys start with amcp_ and the secret is returned once, at creation.

# Create a key (signed in to Aparok)
curl -X POST https://aparok.com/api/mcp-keys \
  -H "Content-Type: application/json" \
  -d '{"name":"Claude Code"}'

# Claude Code
claude mcp add --transport http aparok \
  https://aparok.com/api/mcp \
  --header "Authorization: Bearer amcp_your_key"

// mcp.json — Cursor and similar clients
{
  "mcpServers": {
    "aparok": {
      "url": "https://aparok.com/api/mcp",
      "headers": {
        "Authorization": "Bearer amcp_your_key"
      }
    }
  }
}

The toolset · read

Eleven read-only tools. No credits, no side effects.

These never refresh data or spend anything — they report what Aparok has already verified. Every call starts from a workspaceId returned by list_workspaces.

list_workspacesRead

List workspaces

Lists every Aparok workspace the connection can reach, along with tracked domains and prompt sets. Start here — every other tool needs a workspaceId from this call.

Inputsno parameters

“Which Aparok workspaces can you see?”

get_visibility_summaryRead

AI visibility summary

Compares verified mentions, citations, AI sessions, conversions, and open issues for a period against the period before it — with the evidence counts behind each rate.

InputsworkspaceId · days (1–365, default 7)

“How did our AI visibility change over the last 30 days versus the 30 before?”

get_prompt_performanceRead

Buyer prompt performance

Breaks down tracked buyer prompts across assistants: mention rate, citation rate, which prompts you are losing, who wins them instead, and an excerpt of the actual answer text.

InputsworkspaceId · days · limit (1–100, default 25)

“Which buyer prompts are we losing this month, and who is being cited instead?”

get_competitor_visibilityRead

Competitor visibility

Ranks the competitors and third-party domains winning mentions and citations, plus the prompt-level gaps where they beat you.

InputsworkspaceId · days

“Rank our competitors by AI citations and show the prompts where they beat us.”

get_ai_trafficRead

AI traffic & conversions

Analyzes attributable AI referral sources, landing pages, intent, engagement, traffic flow, conversions, and period-over-period deltas.

InputsworkspaceId · days

“Which landing pages are getting AI-referred traffic, and do any of them convert?”

get_site_auditRead

Site AI-readiness audit

Returns the latest crawl findings — schema, trust, structure, content, and AI-readiness scores — plus ranked issues and your weakest pages. Reads stored results; never starts a crawl.

InputsworkspaceId

“What are our most critical AI-readiness issues right now?”

get_recommendationsRead

Prioritized recommendations

Returns ranked next actions with impact level, the affected page, the supporting evidence, and a confidence signal for each.

InputsworkspaceId · limit (1–100, default 25)

“Give me the top 10 recommendations and draft a ticket for each.”

list_prospect_analysesRead

List prospect analyses

Browses existing Prospect Fit analyses with pagination, filtered by processing status or by Activate / Review / Skip decision.

InputsworkspaceId · processingStatus? · decision? · limit · offset

“List every prospect we marked Activate in the last batch.”

get_prospect_analysisRead

Get prospect analysis

Retrieves one complete fit assessment: decision, score breakdown, observed evidence, inferred requirements, capability matches, campaign plays, and the outreach angle.

InputsworkspaceId · analysisId

“Pull the full analysis for acme-fintech.com and write me a first-touch email.”

get_prospect_analysis_statusRead

Prospect analysis status

Checks the processing stage of a queued analysis (queued → crawling → extracting → scoring → completed) without pulling the full evidence payload.

InputsworkspaceId · analysisId

“Is that prospect analysis done yet?”

get_prompt_run_statusRead

Prompt run status

Follows a prompt-check batch queued by run_prompt_check — per-provider status, observation counts, and any errors.

InputsworkspaceId · batchId

“Check on the prompt run you started.”

The toolset · write

Four tools that do real work — and say so first.

These queue jobs and draw on your plan, so they require mcp:write and run only when you explicitly ask. Nothing here edits or deletes existing data.

run_prompt_checkUses credits

Run a fresh prompt check

Queues a fresh execution of an approved prompt watchlist across its configured assistants and returns a batch ID. Consumes prompt-run credits.

InputsworkspaceId · promptSetId

“Run our approved prompt set now and tell me when it finishes.”

run_site_auditUses credits

Run a fresh site audit

Starts a new crawl and AI-readiness audit, optionally a full-site rescan, and returns a job ID. Consumes scan capacity.

InputsworkspaceId · fullSiteRescan (default false)

“Kick off a fresh full-site audit for our domain.”

generate_strategy_reportUses credits

Generate strategy report

Generates or refreshes the strategy report and executive AI-visibility briefing from verified workspace data, and returns a link to it.

InputsworkspaceId

“Refresh our strategy report and summarize what changed.”

analyze_prospect_fitUses credits

Analyze prospect fit

Scores up to 20 prospect domains against your seller profile. Returns cached Activate / Review / Skip decisions instantly and queues fresh analyses where needed. Fresh runs consume credits.

InputsworkspaceId · prospects[] (1–20 domains) · forceRefresh

“Qualify these 12 domains from my list and tell me which ones to work.”

How teams use it

Four conversations worth having on a Monday.

Workflow 01

Monday visibility review

“Summarize our AI visibility for the last 7 days, show the prompts we're losing, and rank the competitors beating us.” The assistant chains get_visibility_summary, get_prompt_performance, and get_competitor_visibility, then writes the update for you.

Workflow 02

Turn findings into a work plan

“Pull our site audit and the top recommendations, then group them into a sprint by effort.” get_site_audit and get_recommendations return the evidence and affected pages, so the plan cites real URLs instead of guesses.

Workflow 03

Refresh data mid-conversation

“Run the prompt check and let me know when it's done.” run_prompt_check queues the batch and returns a batch ID; the assistant polls get_prompt_run_status until every provider run completes.

Workflow 04

Qualify a prospect list in chat

“Score these domains and draft outreach for the Activate ones.” analyze_prospect_fit returns cached decisions immediately, get_prospect_analysis supplies the evidence, and the assistant writes the message.

Access & permissions

The connector can only reach what your account can already open.

Workspace access is re-checked on every single tool call — a connection can never read a workspace you are not a member of. Scopes decide how far it goes beyond reading, and you can revoke either credential at any time.

ChatGPT AppsClaude connectorsClaude CodeCursorMCP SDK clients

What each credential grantsPick OAuth for chat clients, an API key for anything headless.

mcp:readAll 11 read-only tools

mcp:writeRequired for the 4 queueing tools

offline_accessRefresh without re-authorizing

MCP API keyYour full account access, one bearer token

Workspace service keyLocked to a single workspace

Token lifetimesAccess tokens last 1 hour and refresh tokens 30 days. A revoked credential fails on its very next call.

FAQ

Questions teams ask about the Aparok MCP server.

What is the Aparok MCP server?

It is a remote Model Context Protocol (MCP) server at https://aparok.com/api/mcp that exposes 15 Aparok tools — AI visibility summaries, buyer-prompt performance, competitor citations, AI traffic, site audits, recommendations, and Prospect Fit qualification — to any MCP-compatible AI assistant. Once connected, you query your Aparok data in plain language instead of opening the dashboard.

Which assistants can connect to it?

Anything that speaks MCP. That includes ChatGPT — via Apps & Connectors, the feature formerly known as plugins — Claude connectors on web, desktop, and mobile, Claude Code, Cursor, and any custom client built on the MCP SDK. The server implements the streamable HTTP transport with OAuth 2.1 discovery, so most clients need only the URL.

How does authentication work?

Two ways. OAuth 2.1 with PKCE and dynamic client registration is the default for ChatGPT and Claude — you paste the URL, get redirected to Aparok, approve, and you're connected; no key handling at all. For headless clients that prefer a static credential, mint an MCP API key from the /api/mcp-keys endpoint and send it as a Bearer token. Existing workspace service keys also work and stay scoped to their single workspace.

What can the connection actually access?

Only the workspaces your Aparok account can already open — the server re-checks workspace access on every tool call. OAuth grants carry scopes: mcp:read allows all eleven read tools, and mcp:write is required for the four tools that queue work or spend credits. A workspace service key exposes exactly one workspace.

Which tools cost credits?

All eleven read tools are free and never refresh data. Four tools do real work and draw on your plan: run_prompt_check uses prompt-run credits, run_site_audit uses scan capacity, analyze_prospect_fit charges 3 credits per fresh account (cached results within 30 days are free), and generate_strategy_report regenerates your report. Assistants are instructed to run these only when you explicitly ask.

Can the assistant change or delete my data?

No. There is no tool that edits or deletes anything. The four write tools only queue new work — a prompt run, a crawl, a report regeneration, or a prospect analysis. Everything else is strictly read-only and annotated as such so clients can label it in the approval prompt.

Do I need a paid plan?

You need an active Aparok plan, and the 7-day free trial counts. The MCP server is included at no extra cost on every plan — you are only ever spending the same credits the dashboard would.

How do I revoke access?

Revoke the credential at its source: OAuth grants at the /oauth/revoke endpoint, MCP API keys by calling DELETE on /api/mcp-keys. Revocation is immediate — the next tool call from that client fails authorization. Removing the connector inside ChatGPT or Claude stops it being used from that client, but revoking in Aparok is the authoritative kill switch.