Overview
The Mandelbaum Analytics MCP lets supported AI assistants answer questions about the search performance of projects assigned to your Mandelbaum account. Authentication happens through the same account you use for the Mandelbaum dashboard.
The analytics tools are read-only. They can report search volume, click-through rate, popular and underperforming queries, and product conversion funnels, but they cannot change project settings or analytics data.
- MCP server URL
- https://backend.mandelbaum.ai/mcp
- Name
- Mandelbaum Analytics
- Transport
- Streamable HTTP
- Authentication
- OAuth 2.1 with browser login
- API key
- Not required
- OAuth client ID or secret
- Not required
Before you connect
- A Mandelbaum account
- At least one project assigned to that account
- Search analytics collected for the period you want to inspect
- A client that supports remote Streamable HTTP MCP servers and OAuth
Connect a client
Choose your client below. Every client uses the same production endpoint and opens Mandelbaum in your browser to complete OAuth authentication.
Claude
These steps apply to Claude on the web and to remote connectors in Claude Desktop and the Claude mobile apps.
- Open Claude and go to Customize → Connectors
- Select + → Add custom connector
- Enter Mandelbaum Analytics as the name and https://backend.mandelbaum.ai/mcp as the remote MCP server URL; leave the optional client ID and client secret empty
- Select Add, then Connect, and complete the Mandelbaum login in your browser
- In a conversation, select + next to the composer, open Connectors, and enable Mandelbaum Analytics
In Team and Enterprise workspaces, an Owner or Primary Owner must add the connector under Organization settings first. Each member then authenticates with their own Mandelbaum account and sees only their assigned projects.
Claude Code
Add the remote server with user scope to make it available in every Claude Code project.
bash
claude mcp add --transport http --scope user \
mandelbaum-analytics https://backend.mandelbaum.ai/mcp- Run the add command in a terminal
- Start Claude Code and enter /mcp
- Select mandelbaum-analytics and complete the browser login
- Use the verification command to confirm the server is connected
Verify the connection
claude mcp listChatGPT
ChatGPT connects to remote MCP servers through developer mode and custom apps. Availability depends on your plan and workspace controls.
- Enable Developer mode in ChatGPT settings
- Open Settings → Plugins and create a developer-mode app; managed workspaces may require an administrator to create it
- Use Mandelbaum Analytics as the name, describe it as read-only search analytics, and enter https://backend.mandelbaum.ai/mcp as the MCP server URL
- Select OAuth and leave the client ID and client secret empty
- Create the app or select Scan Tools, then complete the Mandelbaum browser login
- Start a new chat, select + near the composer, choose More, and enable Mandelbaum Analytics
If these options are unavailable, ask your workspace administrator to enable developer mode or custom apps. After server tools change, use Refresh or Scan Tools again.
Codex
Add the Streamable HTTP server and start the OAuth login from the Codex CLI.
bash
codex mcp add mandelbaum-analytics --url https://backend.mandelbaum.ai/mcp
codex mcp login mandelbaum-analytics- Run the add command
- Run the login command
- Complete the Mandelbaum login in your browser
- Use the verification command or open /mcp inside Codex
Verify the connection
codex mcp listThe ChatGPT desktop app, Codex CLI, and Codex IDE extension share the same local MCP configuration.
Cursor
Add Mandelbaum Analytics to Cursor's global MCP configuration so it is available across projects.
json
{
"mcpServers": {
"mandelbaum-analytics": {
"url": "https://backend.mandelbaum.ai/mcp"
}
}
}- Add the JSON configuration to ~/.cursor/mcp.json
- Restart Cursor and open its MCP settings
- Authenticate mandelbaum-analytics when prompted
- Cursor Agent CLI users can run the login and list-tools commands
Verify the connection
cursor-agent mcp login mandelbaum-analytics
cursor-agent mcp list-tools mandelbaum-analyticsOther MCP clients
Any client with remote Streamable HTTP and OAuth support can use the same connection details.
text
Name: Mandelbaum Analytics
Transport: Streamable HTTP
URL: https://backend.mandelbaum.ai/mcp
Authentication: OAuth- Create a remote MCP server connection
- Enter the displayed name, transport, URL, and authentication method
- Leave optional OAuth client credentials empty
- Let the client discover OAuth and complete the Mandelbaum browser login
Clients that support only local stdio servers, static bearer tokens, or legacy SSE without OAuth cannot connect directly. Use the client's remote MCP and OAuth feature instead of an API-key header.
Available analytics tools
The assistant chooses these tools from your natural-language request. All analytics tools respect the projects assigned to the authenticated account.
- list-projects
- Projects available to the account, including the domain used in other calls.
- get-search-volume
- Searches, product clicks, and click-through rate for a period.
- get-top-search-queries
- Up to 20 queries with the highest search volume.
- get-low-performing-queries
- Up to 20 frequently searched queries with low product-click CTR.
- get-top-products
- Up to 10 products with the most search-originated clicks or conversions.
- get-products-breakdown
- The click, cart-add, and checkout funnel for up to 20 top-converting products.
The same MCP server also exposes ecommerce-search, get-locales-for-shop, and get-facets-for-search-query. These are product search tools, so seeing them alongside the analytics tools is expected.
Projects, periods, and filters
Describe the result you want in plain language. The assistant translates your request into these supported arguments.
Select a project
If your account has one project, its domain is optional. With multiple projects, ask the assistant to list them first and name the domain in later questions. Domains work with or without https:// and tolerate differences in the www. prefix.
List my Mandelbaum projects, then show search volume for shop.example.com.
Select a period
Date boundaries are calculated in UTC. Custom ranges use inclusive start and end dates, and the start date must be earlier than the end date.
- week
- The current day and previous six days.
- month
- The previous month through now; this is the default.
- quarter
- The previous 90 days through now.
- year
- The current calendar year through now.
- custom
- An inclusive range using startDate and endDate in YYYY-MM-DD format.
Filter search types
Analytics includes every search type by default. Ask for one or more of these values when you need a narrower result.
- REGULAR
- Ordinary text searches.
- SKU_EAN
- SKU or EAN searches.
- TOP_PRODUCTS
- Automatically displayed top products.
- TOP_QUERY
- Searches initiated from a suggested top query.
Example prompts
Ask questions naturally. The assistant can call several tools when it needs to compare periods or provide a broader analysis.
“List the projects available to my Mandelbaum account.”
“How many searches and product clicks did shop.example.com receive in the last month, and what was the CTR?”
“What were the 20 most popular search queries for shop.example.com this quarter?”
“Which regular searches had the lowest CTR last month? Include searches, clicks, and CTR.”
“Show the top products from search and their click-to-cart-to-checkout funnel.”
“Compare search volume and CTR for shop.example.com between 2026-05-01–2026-05-31 and 2026-06-01–2026-06-30.”
“Summarize the most important search-performance problems and suggest which queries or products I should investigate first.”
How metrics are interpreted
These definitions keep results consistent across clients and analytics tools.
- Searches
- Search sessions that returned results, as defined by Mandelbaum's analytics service.
- Clicks
- Product-detail-page clicks attributed to search.
- CTR
- Product clicks divided by searches, multiplied by 100 and rounded to two decimal places.
- Top products
- Products ranked by attributed conversions. The detailed funnel separates product discovery, clicks, cart additions, and checkouts.
Troubleshooting
Most connection problems come from using the wrong transport or trying to authenticate with an API key.
The client returns 401 Unauthorized
The MCP does not accept your normal Mandelbaum API key. Open the client's MCP settings and start or repeat the OAuth browser login. Confirm that the URL ends in /mcp.
The client asks for an OAuth client ID and secret
Leave both fields empty unless a managed corporate client explicitly requires them. The server supports dynamic client registration, so supported clients register themselves.
The connection succeeds but no tools appear
Confirm that the transport is Streamable HTTP, refresh or rescan the tools, and restart local clients. Also check whether a workspace administrator disabled custom MCP tools.
The tool says that multiple projects are available
Ask the assistant to call list-projects, choose a returned domain, and include that domain in your next analytics question.
An expected project is missing
The authenticated account does not have access to that project, or you signed in with another account. Disconnect, reconnect, and verify the account used in the browser login.
A custom period is rejected
Use YYYY-MM-DD for both dates, set the period to custom, and make sure the start date is before the end date.
The result is empty
Try a wider period and remove search-type filters. If the result stays empty, confirm that the project collected search and conversion events during that period.
Security and privacy
The connection is read-only, but analytics results still contain your business data. Connect clients deliberately and review their data policies.
- Connect only through https://backend.mandelbaum.ai/mcp
- OAuth tokens are issued to the MCP client; the assistant never needs your Mandelbaum password or API key
- Tool results are sent to the AI client you choose, so review that provider's data handling and your organization's policies
- Access follows the projects assigned to your Mandelbaum account; removing project access prevents later analytics calls for that project
- To remove a client's access, disconnect or remove its MCP server and clear the client's stored authentication