AI & Analysis
Ask your feedback anything, from the AI you already use.
Connect Claude, ChatGPT, Codex or your own agents to Formbricks over MCP. Your CX team asks a question in plain English and gets the answer in seconds, with the comments behind it. Your IT team approves one connection: everyone signs in with their own login, and an agent never does more than the person who connected it.
Pick a time right in the form.
What are customers saying about delivery? Is anything holding us back?
Used 3 Formbricks tools
Delivery is one of your strengths. 39 of the 48 comments about it are positive, and most praise parcels arriving early. The one request that keeps coming back is choosing a delivery slot. I found 11 more comments like it.
“Parcel came a day early and was packed with care.”
App reviewVery positivematch 0.91
“Delivery was faster than the shop promised. Great service.”
Post-purchase surveyPositivematch 0.88
“Love the speed, but I'd like to pick a delivery slot.”
Support ticketMixedmatch 0.84
list_feedback_datasets
Customers Europe · 7,667 records
search_feedback_records
query: "delivery and parcel arrival" minScore: 0.5
48 matches, best 0.91
find_similar_feedback_records
feedbackRecordId: "fr_2c91" // "Love the speed, but…"
11 similar comments
3 read calls, as Anna · Northwind EU
Semantic search
Answers you can check.
The agent searches your experience data by meaning, so “easy returns” finds the customer who sent a jacket back from their phone. It narrows down by source, question, language, date, sentiment or emotion, and quotes the comments behind every answer, so your team can trust what it repeats in the next meeting.
When the question is how many, the agent asks for a count. Counts come back as a number, without the comments, so the AI tool sees only what the question needs.
“Sent the jacket back from my phone in two minutes.”
0.89“My refund arrived before the parcel reached the warehouse.”
0.83“The return label was already in the box. So simple.”
0.80
Matched by meaning, not by keyword.
- Sentiment: Very positive
- Emotion: Joy
- Last 30 days
214
records, returned as a number. No comment leaves Formbricks.
Filter by
- Source
- Question
- Language
- Date range
- Sentiment
- Emotions
- End user
Use cases
Hand the busywork to an agent.
With write access, agents add feedback records, draft surveys and set up workflows, so a request in chat becomes finished work in Formbricks. New workflows always start as drafts and can be tested without sending anything, so a person decides what goes live.
Pick a use case to see the tool calls behind it.
Pick a use case
A customer says our size guide runs small for jackets. Is this a one-off?
- search_feedback_recordsFound the comment, match 0.93
- find_similar_feedback_records23 similar comments
Not a one-off. 23 similar comments say the same, from app reviews, support tickets and the post-purchase survey. Most of them mention the sleeves.
Setup
Connected in a minute.
Add your Formbricks URL to Claude Code, the Claude apps, ChatGPT or Codex, sign in and approve. The client registers itself, so nobody files a ticket for a key. Other MCP clients with OAuth connect the same way when run locally, and self-hosters can allow more. Clients without OAuth can use an API key instead.
Rolling it out to a team? A Claude Code project file holds the server URL and no credentials, and every developer signs in on first use.
1Add the URL
Name
Formbricks
Server URL
cx.northwind.com/api/mcp
No API key to paste
2Sign in as yourself
3Approve access
- Read and edit surveys
- Read and edit workflows
- Read and write feedback records
Permissions
An agent never does more than you can.
Each connection carries the scopes its person approved, read or write for surveys, workflows, feedback records and responses. Every call is then checked against that person's role in the workspace. Connect with read scopes only, and reading is all the agent can do.
Every connection is listed under Authorized Apps with its scopes and approval date, and revoking one cuts it off at once. Every change lands in the audit log, refused attempts included.
May
- Search, count and read feedback records
- Add feedback records, up to 50 per call
- Create and edit surveys
- Create workflows, always as drafts
May not
- Read survey responsesScope not approved
- Correct or delete recordsNeeds an owner or manager
- Change the workspace CSSNeeds an owner or manager
- Claude6 scopes · Oct 21Revoke
- ChatGPT2 scopes, read only · Oct 9Revoke
- Codex2 scopes, read only · Sep 30Revoke
- 10:42Claudecreate_workflowSuccess
- 10:47Claudecreate_surveySuccess
- 11:05Codexpatch_surveyRefused
Your login, not a shared key
Clients connect over OAuth 2.1 and each person signs in with their own Formbricks account. There's no API key to hand out, paste or rotate.Short-lived tokens
Access tokens last 15 minutes and refresh quietly for up to 30 days. Write access is confirmed again from time to time.Revoke in one click
Every authorized app sits under Settings, Account, Authorized Apps. Click Revoke and its access ends at once, until someone authorizes it again.Rolled out by your admins
On Claude Team and Enterprise, an organization owner adds the Formbricks connector once. Each member then connects with their own login.
For IT
On the AI your company already approved.
Formbricks works with the assistant your teams already use, so there's no new AI vendor to review. What an agent reads goes to that tool under the contract you already have, and everything else stays in your infrastructure.
The MCP server is part of Formbricks, with its OAuth provider built in. Self-hosted, it runs on your own domain, and hosted assistants such as claude.ai and ChatGPT reach it over public HTTPS.
The AI tool you approved
Gets what the agent asks for, under the contract you already have with that vendor.
- Claude
- ChatGPT
- Codex
- Your agents
Formbricks MCP server
Part of Formbricks, with its OAuth provider built in. Uses the same APIs and checks as the app.
- OAuth 2.1
- Scope check
- Role check
- Audit log
Your infrastructure
Feedback records stay in Formbricks Hub, on your own PostgreSQL.
- Formbricks Hub
- PostgreSQL
- Your embedding model
Customer stories
Trusted by teams that own their data
See how public sector, research and product teams run feedback on Formbricks.
Public Sector
How DigitalService runs user research for German ministries on self-hosted Formbricks
- 2 weeks
- From decision to production
- >15
- Ministry projects
- 0h
- Manual compliance work
Read full story“Self-hosting makes us faster. Because we work on a project basis with various ministries, it is our responsibility to respect their rigorous standards for data protection.”
Joshua Nowak
User Researcher @ DigitalService
Market Research
How Borderless Access went from months to minutes, with zero engineering time per survey
- 15 days
- Implementation
- Mins
- To launch a new survey
- 0 hours
- Engineering time
Read full story“The rule with Formbricks was simple: do the integration once, and then leave it. My engineering team spent about 15 days getting everything set up, and since then marketing has been completely self-sufficient.”
Srihari Udugani
Director of Technology @ Borderless Access
Market Research
How Opinodo gathers millions of responses with a self-hosted instance of Formbricks
- 38%
- Cost reduction
- 8%
- Higher completion rate
Read full story“Big Formbricks fan here! It has enabled us to radically improve our customer retention and even exploring new revenue streams.”
Rasmus Lønne
Managing Director @ Opinodo
Frequently asked questions
Which license includes it?
What does semantic search need?
Which ChatGPT plans work?
Where does it run?
- Read the code
- ISO 27001 certificate on request
- SOC 2 Type II report on request
- Accessibility report available
Ask your own feedback a question.
Pick the AI tool your team already uses and the first question it should answer. We'll connect it to Formbricks and ask it together.