ai-shelf connector for Claude and ChatGPT
Last updated: 28 October 2026
Use your ai-shelf results inside Claude, ChatGPT and other AI assistants: setup, tools, data handling and limits.
1. What the connector does
The ai-shelf connector brings your ai-shelf results into Claude, ChatGPT and other AI assistants that support the Model Context Protocol (MCP). Ask in plain language how often AI assistants name your brand, who they recommend instead, which websites they cite and what to fix first, and the assistant answers from your published ai-shelf results.
It reads the same results you see in the ai-shelf app, for your own account only. It can also send Markapp a request for a new run or an AI SEO check; nothing runs or is charged until Markapp confirms.
2. Connection details
| Setting | Value |
|---|---|
| Server URL | https://aishelf.tech/mcp |
| Transport | Streamable HTTP (MCP) |
| Authentication | OAuth 2.1 with PKCE and Dynamic Client Registration. You sign in with your ai-shelf email and password on aishelf.tech. |
| Who can use it | People with an ai-shelf client account. Accounts are created by Markapp by invitation; book a demo to get one. |
3. Set up in Claude
- Claude (web, desktop and mobile): open Settings, then Connectors. Find ai-shelf in the directory and choose Connect, or choose Add custom connector and enter the server URL
https://aishelf.tech/mcp. Sign in with your ai-shelf account and choose Allow. - On Team and Enterprise plans an owner may need to add the connector for the organisation first.
- Claude Code: run
claude mcp add --transport http ai-shelf https://aishelf.tech/mcp, then/mcpin Claude Code to sign in. - In a chat, turn the connector on from the tools menu if it is off, then ask, for example: “How visible is my brand in AI answers this month?”
4. Set up in ChatGPT and Codex
- ChatGPT: open Settings, then Apps. Find ai-shelf and choose Connect, sign in with your ai-shelf account and choose Allow. Before the app is listed, a workspace admin can add it with developer mode: create an app with the server URL
https://aishelf.tech/mcpand OAuth authentication. - Codex: run
codex mcp add ai-shelf --url https://aishelf.tech/mcp, thencodex mcp login ai-shelfto sign in. - In a chat, ask for ai-shelf by name the first time, for example: “Use ai-shelf to show the top fixes for my brand.”
5. Tool reference
| Tool | Title | Type | Inputs | What it returns |
|---|---|---|---|---|
list_brands | List my brands | Read | none | Your account’s brands and markets, what your package includes (runs and AI SEO checks left), the latest results, a run in progress, earlier runs and open requests. |
get_visibility_score | Get AI visibility score | Read | brand, scan_id (optional) | AI Visibility Score (0 to 100), rank against competitors, mention rate, top-3 rate, share of answer, average position, own-site citation rate, the same per AI assistant, and the trend over runs. |
get_competitors | Get competitor ranking | Read | brand, scan_id (optional) | The market ranking, the top rivals and why AI assistants name them, the topics each one leads, and the gaps where a rival is named instead of you. |
get_top_fixes | Get top fixes | Read | brand, scan_id (optional), limit (1 to 10) | The evidence-based fixes that would most raise your share of AI answers, most important first, each with the answer IDs behind it. |
get_cited_sources | Get cited sources | Read | brand, scan_id (optional), limit (1 to 30) | The websites AI answers cite, split into your own site, competitors’ sites and third parties, with how many answers name you or only competitors. |
request_new_run | Request a new run | Write (request only) | brand, message (optional), add_assistants (optional) | Sends Markapp a request for a new run of your panel, optionally on more AI assistants. Nothing runs or is charged until Markapp confirms. |
request_ai_seo_check | Request an AI SEO check | Write (request only) | brand, page_url, focus (optional) | Sends Markapp a request for an AI SEO check of one page of your website. Nothing runs or is charged until Markapp confirms. |
Read tools are marked read-only. The two write tools only create a request, the same request as the buttons in the ai-shelf app; they never start a run, change your package or delete anything.
Scores run from 0 to 100. Rates and shares are fractions from 0 to 1. Results marked sample data come from a demo account, not from real AI answers.
6. Example prompts
- “List my brands in ai-shelf and their latest AI visibility scores.”
- “Who do AI assistants recommend instead of us, and for which topics?”
- “Show the top 5 fixes with the answer IDs behind them.”
- “Which websites do AI answers cite for our category, and where are we missing?”
- “Ask Markapp for a new run next month that adds Gemini.”
7. Data and security
- You sign in on aishelf.tech; the AI assistant never sees your password. Access tokens are signed, expire after 2 hours, and stop working at once when you change your password or your account is closed.
- The connector returns only your own account’s results: never other companies’ data, API costs or Markapp’s internal notes. Markapp staff accounts see only Markapp’s own account through the connector.
- We receive the tool calls and their inputs, not your conversation. See the Privacy policy (section 5) and the Terms of service.
- To disconnect, remove ai-shelf in your assistant’s settings, or write to us to end every connection of your account.
8. Method and limits
ai-shelf asks each AI assistant the same customer questions through its official API with web search on: a consistent measurement panel, not a copy of what any one person sees on screen. Answers vary from day to day, so full runs repeat each question and report the median.
Pricing is per run, with no subscription: USD 1,000 for the first run and USD 500 for each further run, for 2 AI assistants. Requests from the connector are priced the same way and confirmed by Markapp first.
9. Support
Questions, problems or a data request: adtech@markappmedia.site. See the support page.