MCP server
An AI visibility MCP server for Claude Code and Cursor
Connect your AI assistant to Rankry and ask about your brand’s visibility, competitors, cited sources, Search Console data and AI traffic in plain words. One API key, 21 tools, and every tool only reads, except one that runs a PageSpeed audit.
7-day Base trial · No card needed
Claude Code · rankry
4 tool calls- get_brand_summary: UGC Planner, last 30 days
- get_competitors: Notion, Collabstr, Brkaway
- list_sources: 3 domains cite rivals, not you
- get_gsc_performance: top queries, 28 days
Sample data
In short
The Rankry MCP server is a Model Context Protocol endpoint that lets AI tools such as Claude Code, Cursor and compatible MCP clients read your Rankry data. You create an API key in Rankry, add it to your client as a Bearer token, and the client gets 21 tools covering visibility, sentiment, competitors, prompts, sources, history, AI traffic, Search Console, PageSpeed and tasks. All of them are read-only except running a PageSpeed audit.
The basics
What is an MCP server for AI visibility?
MCP, the Model Context Protocol, is an open standard for connecting AI assistants to outside tools and data. An MCP server publishes a list of tools. An MCP client, such as Claude Code or Cursor, decides when to call them while it works on your request, and reads the results back into the conversation.
Rankry’s MCP server publishes your AI visibility data as tools. Ask your assistant “How is UGC Planner doing in AI search this month?” and it can call get_brand_summary, read the numbers, and answer in the chat. It can also combine Rankry data with whatever else it has open: your codebase, your content drafts, or another server’s data.
It is the same data you see in the Rankry app, reached from where you already work. Use it to write a status update, check a page before rewriting it, or build your own reporting agent on top of Rankry.
21 tools
Which tools does the Rankry MCP server include?
Most tools take an optional project ID, so one key works across all the brands on your account. Your client sees the tool names and descriptions and picks the right ones.
| Tool | What it returns | Access |
|---|---|---|
| list_projects | The brands (projects) on your account, each with its Rankry Score and last report date | Read |
| list_reports | Reports for a brand, newest first, with score, status and dates | Read |
| get_brand_summary | Headline summary: Rankry Score and its change, visibility per engine, sentiment and top insights | Read |
| get_visibility | Visibility per engine, prompts analysed, and which categories are strong, weak or invisible | Read |
| get_sentiment | Sentiment per engine: how answers that name you split between positive, neutral and negative | Read |
| get_competitors | A leaderboard of you and your tracked competitors with visibility, average position and sentiment | Read |
| get_insights | The rule-based insights from your latest report, each with a severity and a status | Read |
| get_history | Rankry Score and per-engine visibility, sentiment, position and diversity over 7, 30, 90 or 365 days | Read |
| get_top_prompts | Prompts from a report ranked by how you perform: engines that named you, rank, sentiment, rivals named | Read |
| list_sources | Domains AI cited in your space, with their type and status (yours, gap or mixed) | Read |
| get_ai_traffic | AI referral visits: totals, platforms, landing pages, queries, countries, devices and engagement | Read · Pro |
| get_integrations | Which integrations your plan includes and whether each one is connected | Read |
| get_gsc_performance | Search Console clicks, impressions, CTR and position by query, page, country, device, date or hour | Read · Search Console |
| get_gsc_index_status | URL inspection for up to 25 URLs: index status, coverage, robots.txt verdict, last crawl, sitemaps | Read · Search Console |
| get_gsc_sitemaps | Submitted sitemaps with status, error and warning counts, and last download time | Read · Search Console |
| get_pagespeed_reports | Saved PageSpeed results: Lighthouse scores and lab metrics for each URL and device | Read · PageSpeed |
| get_pagespeed_report | One full PageSpeed report: scores, Core Web Vitals, field data and the top failing audits | Read · PageSpeed |
| run_pagespeed_audit | Runs a fresh PageSpeed Insights test for a URL and saves the report | Runs and saves · PageSpeed |
| list_tasks | Your tasks with status, priority and type | Read |
| get_task | Full detail of one task: summary, brief, checklist and notes | Read |
| get_guide | Rankry’s own how-to guides, such as connecting Search Console or reading your score | Read |
Tools marked Pro need AI Traffic on your plan and the script installed. Tools marked Search Console or PageSpeed need that integration on your plan and connected.
Claude Code
How do I connect Claude Code to Rankry?
Claude Code has its own tab on Rankry’s MCP page, with the full command ready to copy.
- 01
Open the MCP server page
In your project, go to Integrations and open “MCP server”. It shows the server URL and your API keys.
- 02
Create an API key
Name it after the tool you’ll use it in. The key starts with rnk_ and is shown once, so copy it before you close the box.
- 03
Copy the Claude Code command
On the “Claude Code” tab, the command reads: claude mcp add --transport http rankry <server URL> --header "Authorization: Bearer rnk_…"
- 04
Run it in a terminal
Claude Code registers the server as “rankry”. Start a session and Claude can answer questions about your Rankry data.
- 05
Ask a first question
Try “Which prompts am I losing to competitors?” or “Where are competitors cited but I am not?” Claude picks the tools it needs.
Other clients
Can I use Rankry MCP with Cursor and compatible clients?
Rankry gives two ready setups: a config-file entry and a Claude Code command. Any client that can reach a remote HTTP MCP server and send an Authorization header can use the same details.
Cursor and config-file clients
The “Config file” tab gives a JSON mcpServers entry named “rankry” with type http, the server URL and an Authorization header carrying your key. Use it in Cursor, in a project’s .mcp.json, or in any client that reads an MCP config file.
ChatGPT and other OAuth-based connections
The ChatGPT app expects an OAuth account connection for private MCP data. Rankry currently authenticates with an API key and does not provide that OAuth flow. Use Claude Code, Cursor or another client that supports a custom Authorization header. Other apps must be checked against their own authentication requirements.
Local-only clients
The server speaks streamable HTTP only, not stdio or SSE. A client that can only launch local servers needs a bridge that forwards its requests to a remote HTTP server.
Agents and scripts you build
Any MCP SDK can connect with the server URL and the Bearer header. The server is stateless and answers HTTP POST requests with JSON, which keeps it simple to call from your own code.
Example prompts
What can I ask my AI assistant about my visibility?
Ten questions that work well, and the tools a client typically calls to answer them.
“How is my brand doing in AI search this month?”
get_brand_summary with a 30-day period: score, change, visibility per engine and the top insights.
“Which competitor gained the most ground?”
get_competitors for the leaderboard and get_history for the trend behind it.
“Which prompts am I losing, and to whom?”
get_top_prompts lists each prompt with the engines that named you, your rank and the competitors named.
“Where are competitors cited but I am not?”
list_sources returns cited domains with a gap status: they cite rivals, not you.
“Compare this report with the previous one.”
list_reports finds both runs, then get_top_prompts reads each so the client can compare them.
“Which of these 20 URLs are indexed?”
get_gsc_index_status inspects up to 25 URLs in a single call.
“Which queries lost clicks over the last 3 months?”
get_gsc_performance groups Search Console data by query and date in one request.
“Why is our pricing page slow on mobile?”
get_pagespeed_report reads the saved report; run_pagespeed_audit runs a fresh test.
“How much traffic did ChatGPT send us?”
get_ai_traffic, on Pro with the AI Traffic script installed.
“What should I work on next?”
list_tasks and get_task read the tasks you collected from the Action plan.
Permissions
Is the Rankry MCP server read-only?
Almost entirely. Twenty of the 21 tools only read data, and they are marked read-only for clients that check. The exception is run_pagespeed_audit, which asks Google PageSpeed Insights to test a URL and saves the report in Rankry. It needs the PageSpeed integration and your own Google API key, and a repeat call within 60 seconds returns the cached result.
No tool can add or delete prompts, change competitors, start a report, create or complete tasks, or change anything in Search Console. The Search Console access behind the tools is read-only on Google’s side as well.
When you ask your client for Search Console data, Rankry returns the result to that client, and only because you asked. Your client then processes it under its own terms and privacy policy, so review those before connecting.
Security
How are Rankry MCP keys secured?
Keys are simple to create and simple to revoke. These are the rules behind them.
- Authentication is a Bearer API key, sent in the Authorization header
- Keys start with rnk_, are stored hashed, and are shown only once
- Up to 20 active keys per user: make one for each AI tool you connect
- One key reads every project on your account
- Revoke a key on the MCP page and the tool using it loses access
- Rate limits: 60 requests a minute per key and 600 per IP address
- Tools follow your plan: AI Traffic needs Pro; Search Console and PageSpeed tools need those integrations
Explore Rankry
Related features
Which plans include the Rankry MCP server?
Every Rankry subscription can use the MCP server. Individual tools still follow your plan: the AI traffic tool needs Pro, and the Search Console and PageSpeed tools need those integrations included and connected. MCP is not limited to Pro.
Do I need a separate key for each project?
No. One key reads every project on your account. Your client can call list_projects to get each brand’s project ID and pass it to the other tools. It is still a good idea to create one key per AI tool, so you can revoke one without affecting the others.
Does it work with ChatGPT or the Claude desktop app?
The ready-to-copy setups are for Claude Code and config-file clients such as Cursor. The ChatGPT app requires an OAuth account-linking flow for private MCP data; Rankry currently provides API-key authentication instead. Other apps can connect when they support remote HTTP MCP servers with a custom Authorization header.
Does the MCP server support OAuth?
Rankry currently uses a Bearer API key that starts with rnk_. Create a key on the MCP server page, add it to a client that supports Authorization headers, and revoke it when you no longer need the connection. OAuth account linking is not available.
Can my AI assistant change my Rankry data?
No. Twenty of the 21 tools are read-only. The only action is run_pagespeed_audit, which runs Google PageSpeed Insights on a URL and saves the report. No tool can add prompts, edit competitors, start reports, create tasks or touch Search Console settings.
What are the rate limits?
Each key can make 60 requests a minute, and each IP address 600 a minute. Search Console URL inspections also count against Google’s daily allowance for your property, so pass only the URLs you need; the tool accepts up to 25 in one call.
Which transport does the server use?
Streamable HTTP. The server is stateless, accepts POST requests and replies with JSON. It does not offer stdio or SSE transports, so clients that only launch local servers need a bridge to a remote HTTP server.
How is the MCP server different from Rankry Brain?
Rankry Brain is the assistant inside the Rankry app. It uses 19 of the same data tools for the project you are viewing, can draw charts, and never runs anything. The MCP server is for your own AI tools outside Rankry, covers every project on your account, and adds list_projects and run_pagespeed_audit.
Can I use one key for several brands?
Yes. If you track several brands as projects on one Rankry account, one key reads all of them. Ask your client to compare brands or to summarise each one, and it will call list_projects first to find their IDs.
Give your AI tools your AI visibility data
Start a free trial, create an API key, and ask Claude Code or Cursor how your brand shows up in AI answers.
7-day Base trial · No card needed