Connect your AI agent to Feedbaq over MCP

Feedbaq has a remote MCP server. Add it to Claude, ChatGPT, Codex, Cursor, Grok or any other MCP client, sign in to Feedbaq, and choose a workspace. Your agent can then build a study, publish it, and read every response with its transcript.

Server address

https://feedbaq.to/mcp

The server at a glance

Feedbaq MCP server details
Detail	Value
Address	https://feedbaq.to/mcp
Transport	Streamable HTTP
Sign-in	OAuth with PKCE. You approve the connection in your browser; there is no API key.
Access	One workspace, with your own role. The agent can never do more than you can.
Price	Included in the $49 plan. Building and previewing are free; publishing needs the plan.

Set up your agent

Every agent follows the same three steps: add the server address, sign in to Feedbaq, and choose the workspace the agent may use. The exact clicks differ, so pick yours below. The same steps are in Feedbaq under Settings → AI agent.

Claude

Claude app

Open Settings → Connectors in Claude and choose Add custom connector.
Name it Feedbaq, paste the server address below, and add the connector.
Server address
```
https://feedbaq.to/mcp
```
Choose Feedbaq in Search and tools, then select Connect.
Sign in to Feedbaq and choose the workspace you want Claude to use.

Claude Code CLI

Run the command below in a terminal.

Command

claude mcp add --scope user --transport http feedbaq https://feedbaq.to/mcp

Open Claude Code, enter /mcp, choose Feedbaq, and select Authenticate.
Sign in to Feedbaq and choose the workspace you want Claude Code to use.

ChatGPT

Enable Developer mode under Settings → Security and login in ChatGPT.
Open Plugins, choose +, and create a plugin named Feedbaq.
Paste the server address below as its connection URL and review the available tools.
Server address
```
https://feedbaq.to/mcp
```
Add Feedbaq to a conversation, then sign in and choose a workspace.

Developer mode may be controlled by your ChatGPT workspace administrator.

Codex

Codex app

Open Settings → MCP servers in Codex and choose Add server.
Name it Feedbaq, choose Streamable HTTP, and paste the server address below.
Server address
```
https://feedbaq.to/mcp
```
Save the server, restart Codex, and select Authenticate.
Sign in to Feedbaq and choose the workspace you want Codex to use.

Codex CLI

Run both commands below in a terminal.

Commands

codex mcp add feedbaq --url https://feedbaq.to/mcp
codex mcp login feedbaq

Sign in to Feedbaq when the browser opens and choose a workspace.
Restart Codex if it was already open.

Codex shares this connection between its app, CLI, and IDE extension.

Cursor

Cursor editor

Open Customize → MCPs in Cursor.
Add the configuration below to your global MCP file at ~/.cursor/mcp.json.
mcp.json
```
{
  "mcpServers": {
    "feedbaq": { "url": "https://feedbaq.to/mcp" }
  }
}
```
Restart Cursor, open Feedbaq under MCPs, and authenticate.
Sign in to Feedbaq and choose the workspace you want Cursor to use.

Cursor CLI

Add the configuration below to your global MCP file at ~/.cursor/mcp.json.
mcp.json
```
{
  "mcpServers": {
    "feedbaq": { "url": "https://feedbaq.to/mcp" }
  }
}
```
Run agent mcp login feedbaq in a terminal.
Sign in to Feedbaq and choose the workspace you want Cursor to use.

Cursor CLI uses the same MCP configuration as the Cursor editor.

Grok

Grok app

Open grok.com/connectors and choose New Connector → Custom.
Name it Feedbaq and paste the server address below.
Server address
```
https://feedbaq.to/mcp
```
Complete authentication, then choose the Feedbaq workspace you want Grok to use.

On Grok Business or Enterprise, an administrator may need to add the connector first.

Grok CLI

Run the command below in a terminal.

Command

grok mcp add --transport http feedbaq https://feedbaq.to/mcp

Start Grok and use Feedbaq. Grok opens the browser when authentication is needed.
Sign in to Feedbaq and choose the workspace you want Grok to use.

Other agents

Add a remote MCP server and paste the server address below.
Server address
```
https://feedbaq.to/mcp
```
Choose Streamable HTTP as the transport and OAuth as the authentication method if asked.
Sign in to Feedbaq and choose the workspace you want the agent to use.

Check the connection

Start a new conversation and paste this. It only reads, so it is a safe first test.

First prompt

Confirm you can reach my Feedbaq workspace. Then tell me in a few lines what you can do here: create and edit studies, publish them, and read responses and transcripts.

The agent should list your studies, or say there are none yet, and say what it can do. If it says it cannot reach Feedbaq, see Troubleshooting.

What your agent can do

The server has 12 tools. The agent picks them itself; you ask in plain words. Tools that only read are safe to run at any time. The server marks the others as changing data, so agents that ask before acting will ask you first.

The Feedbaq MCP tools
Tool	What it does	Effect
`list_studies`	Lists the studies in the workspace, with links to build, share and read each one.	Reads only
`get_study`	Reads a study's draft, its published versions and its links.	Reads only
`get_study_version`	Reads the exact questions of one published version, the ones a participant saw.	Reads only
`create_study`	Creates a draft study with a welcome page, one empty voice question and a thank-you page.	Edits the draft
`upsert_question`	Adds or changes a question: voice, screen recording, text or choice, with AI follow-ups, skipping, other answer formats and its position.	Edits the draft
`delete_question`	Removes a question from the draft. Published versions do not change.	Edits the draft
`update_study`	Changes the welcome and thank-you pages and the look in the draft. A new name, public link or response limit applies at once, also to a live study.	Draft, and some live settings
`set_image`	Attaches an image or video by URL to a question or a page, or sets the study logo.	Edits the draft
`publish_study`	Publishes the draft as a new version, so the link shows it to participants.	Makes it live
`list_responses`	Lists the responses to a study, with the version each person answered.	Reads only
`get_response`	Reads one full response: every answer, the transcripts with word timings, and links back to the app.	Reads only
`invite_colleague`	Invites a colleague to the workspace by email. Only workspace admins can.	Sends an email

Agents that connected before 16 September 2026 may still use the old names upsert_block and delete_block. They do the same as upsert_question and delete_question.

Prompts to start with

Describe the result you want. Name the people you want to hear from and what you need to decide, and tell the agent whether it may publish. Replace the parts in brackets.

Build a study

Create a short Feedbaq study for people who cancelled in the last 30 days. Ask what was happening when they decided to stop, what they use now, and what would bring them back. Use voice answers with one follow-up each. Don't publish it yet; send me the builder link.

Test a concept

Set up a concept test in Feedbaq for this landing page: [paste the URL]. Ask people to share their screen and say what they think the product does, then what would stop them from signing up. Turn on AI follow-ups.

Read the answers

Read every response to my "[study name]" study in Feedbaq. Give me the three most common reasons, each with two short quotes and the link to the answer they come from.

Compare versions

In my "[study name]" study, compare what people said before and after I published version 2. Did the change to question 3 get clearer answers?

Every template in Feedbaq also has a full prompt behind Use AI agent, written for that study type. For what makes a good study, see the guides on customer discovery questions and concept testing questions, and your AI agent as a researcher.

Access, limits and safety

Your role, one workspace. The agent signs in as you and works in the workspace you chose. It can never do more than your role allows. Only workspace admins can invite colleagues.
Most changes wait for publishing. Most tools edit the draft, and participants see it only after you publish. Publishing freezes the draft as a new version and needs an active plan. Two changes apply at once: a new public link and a response limit, both set with update_study.
The agent cannot reach participants. It cannot send the study link, recruit people or answer for them. You share the link with the people you want to hear from.
The agent cannot delete studies or responses. Its only delete tool removes a question from a draft.
Answer limits. A voice answer can run up to 10 minutes. Screen recordings need a desktop browser and run up to 10 minutes each.
Storage. Recordings, transcripts, typed answers and choices stay as long as the workspace exists.

Disconnect or switch workspace

In Feedbaq, open Settings → AI agent. Under Connected agents, revoke the agent. Its access ends at once. To use another workspace, revoke the connection, then connect again and choose the other workspace when you sign in.

Troubleshooting

The agent says it is not signed in. Repeat the sign-in step in your agent's setup above. In a terminal, run codex mcp login feedbaq for Codex or agent mcp login feedbaq for Cursor CLI.
Claude will not add another connector. Claude's free plan allows one custom connector. Remove one, or use a paid plan.
ChatGPT has no option to add a server. Custom MCP servers need developer mode, a beta on the web for Plus, Pro, Business, Enterprise and Edu. On a company plan, an admin may have to turn it on.
The agent sees the wrong studies. It is connected to another workspace. Revoke it and connect again.
Publishing fails with “subscription required”. Publishing needs an active plan. A workspace owner or admin starts it from the Publish button in Feedbaq. Then ask the agent again.

Still stuck? Write to hello@feedbaq.to with the agent you use and the message it shows.

Questions people ask

Does Feedbaq have an MCP server?

Yes. Feedbaq's remote MCP server is at https://feedbaq.to/mcp. It uses Streamable HTTP and OAuth sign-in, so there is no API key to copy.

Which AI agents work with Feedbaq?

Claude (the app and Claude Code), ChatGPT in developer mode, Codex, Cursor, Grok and any other MCP client that supports remote servers with OAuth.

Does the MCP connection cost extra?

No. It is included in Feedbaq's one plan at $49 a month. You pay for your agent, such as Claude or ChatGPT, with its own provider.

Can my agent read voice and screen recording answers?

Yes, through their transcripts. get_response returns every answer with its transcript, word timings and a link to the moment in the app, so the agent can quote people exactly.

Can my agent publish a study or contact participants?

It can publish a study, which makes the link live and needs an active plan. It cannot send the link to anyone or recruit participants. You share the link yourself.

Updated 1 October 2026. Back to the Help center.