Connecting FeedbackRobot to an AI client

FeedbackRobot can connect to Claude, or any MCP-compatible AI client, so it can work with your feedback alongside everything else you’re doing there.

Start from the AI client, not from FeedbackRobot

This is the part that confuses people: there’s no “connect MCP” button inside FeedbackRobot. The flow starts in the AI client.

In Claude (or your client of choice), add FeedbackRobot as a connector. The client sends you to a FeedbackRobot consent page — that page only works when opened this way, which is why you can’t navigate to it yourself.

Server URL: https://backend-prod.feedbackrobot.com/mcp

That’s all you need. There’s no API key, client ID or secret to generate — leave those fields blank if your client offers them. In Claude that’s Settings → Connectors → Add custom connector; other clients word it differently but the flow is the same.

What happens

1. Continue with email — enter your work email. If you don’t have an account yet this creates one.
2. Verify — via a sign-in link or your password.
3. Choose the organisation — if you belong to more than one.
4. Consent — Connect this app to your org, showing which app is asking and what it will be able to do:
   • Read your feedback, reviews and survey results
   • Connect a new review source by URL
   • Draft and send new surveys
   • Start AI resolutions on feedback

   A connected client can also search your Knowledge Base, so it can answer using your own policies and product detail rather than guessing.

5. Connect, then you’re returned to the app.

Read the scopes before agreeing

Only the first is read-only. The rest all change something, and two of them reach real customers: draft and send new surveys and start AI resolutions can email people on your behalf. Connecting a review source writes to your workspace too, but nobody hears about it.

That’s the point of the integration, but it means an AI client with this connection can email people on your behalf. Grant it only to clients you trust and use yourself.

What it can actually see

A connection is scoped to one organisation, chosen at the consent step. It can’t see your other organisations, and it can’t see anything belonging to another FeedbackRobot customer.

There are 21 tools. Sixteen only read:

reviews and individual review lookups · responses and individual response lookups · surveys · resolutions · touchpoints · review sources · customer lookup, their responses, their conversation history and their sentiment over time · feedback stats · sentiment distribution · resolution rates · search across responses · Knowledge Base search

Five make changes:

What it does	Reaches a customer?
Connect a review source by URL	No
Draft a survey and return a link	No
Send a survey	Yes
Start a resolution	Yes
Send a resolution follow-up	Yes

Every tool tells the client whether it only reads or makes changes, so a well-behaved client warns you before the ones that reach customers. Most ask you to approve each action the first time.

Your role and plan still apply

The connection can’t do anything you couldn’t do yourself in the app.

• Members can read everything, but can’t connect review sources, create or send surveys, or start resolutions.
• Admins and Owners can use all of it.

Actions taken through an AI client count against your organisation’s monthly action limit, exactly like actions taken in the app. Reading is never limited. Sending a survey, drafting a survey and connecting a review source stop and say so once the limit is reached.

Security

• The connection uses OAuth 2.1 with PKCE. The AI client never sees your password.
• Access is scoped to one organisation and to your own permissions.
• We don’t train AI models on your data. Our Privacy Policy covers data handling, sub-processors and retention, including a section on MCP connections specifically.

Disconnecting

There’s currently no way to view or revoke an MCP connection from inside FeedbackRobot. If you need one removed, contact us and we’ll do it.

Removing the connector in your AI client stops that client using it, which is the practical first step.

When something goes wrong

The client says it can’t connect, or that the server needs no authentication. Check the URL is exactly https://backend-prod.feedbackrobot.com/mcp — no trailing slash, no extra path.

A tool says you don’t have permission. Your role doesn’t allow that action. An Admin or Owner in your organisation can do it, or change your role.

A tool says the monthly limit was reached. Your organisation has used its monthly actions. Reading still works; writes resume next cycle, or upgrade your plan.

It connected to the wrong organisation. Remove the connector in your AI client and add it again, then pick the right one at the consent step.

Worth asking it

• “Which customers left negative reviews this month and what did they say?”
• “Draft replies to the open resolutions, matching our usual tone.”
• “What’s our resolution success rate, and do the failures have anything in common?”

The value is combining your feedback with context the assistant already has — your policies, your numbers, the document you’re working on. That’s what it can do that the in-app Assistant can’t.