Hand the feedback to your coding agent
Connect Claude Code, Codex or Cursor to your site's feedback. Your agent reads each thread with its page, anchor and screenshot, and reports back when you approve.
How it works
The uier MCP connector lets your coding tool read feedback and, if you allow it, reply, change status and assign. It cannot edit code, run commands or deploy.
- Connect your coding toolAdd the uier MCP server and sign in. You choose the workspaces, projects and operations it can use.
- Read the feedbackYour agent searches the open threads and reads one with its page, anchor, browser details, discussion and screenshot: the same context as Copy issue.
- Fix it in your codeThe agent proposes the change in your repository. Nothing changes in uier while it works.
- Report backWhen you approve, it replies on the thread as you, labeled with the connection name, and marks the thread done.
Connect your coding tool
Every tool uses one endpoint, https://uier.io/api/mcp. The setup for each is in the app, under Account → Coding agents.
Claude Code
Signs in with your browserAdd uier in your terminal, then run /mcp, choose uier and select Authenticate. uier opens in your browser, where you choose what the tool can reach and approve the connection.
claude mcp add --transport http uier https://uier.io/api/mcpCodex
Access tokenCreate an access token in Account and add uier to ~/.codex/config.toml. The token stays in the UIER_MCP_TOKEN environment variable, not in the file.
Cursor
Access tokenAdd uier to .cursor/mcp.json with the token from the same environment variable.
An access token also works where no browser can sign in, such as SSH sessions, containers and CI. It lasts 30 days and reaches up to 20 projects in one workspace. Other tools are added to Account once their connection has been checked.
What your agent can do
A connection is read-only unless you allow more. You choose its operations when it connects, and they apply to everything it reaches.
| Operation | What it allows |
|---|---|
| Read feedback | List, search and read threads with their context. Always included. Reading does not mark them read. |
| Read screenshots | Open the screenshots attached to feedback. |
| Reply to feedback | Post replies as you, labeled with the connection name. |
| Change status | Mark feedback in progress or done, open it again, and request or cancel a fix check. |
| Assign feedback | Assign feedback to people who can see it, link pull requests to it, and show your coding tool as the agent working on feedback assigned to you. |
The connector cannot edit code, run commands, deploy, delete feedback or confirm a fix. Your coding tool changes code under its own permissions. A retried reply or status change is saved once.
Prompts and a skill
Start a workflow from your coding tool's prompt menu, or save the skill so your agent follows the same steps.
- triage-feedbackTriage feedbackLists and groups the open threads of a project. Changes nothing.
- fix-feedbackPropose a fixReads one thread and proposes a code change in your coding tool. Leaves uier as it was.
- report-backReport backDrafts a reply or a status change and waits for your approval before it writes. Offered when the connection can reply or change status.
Copy or download the uier-feedback skill in Account → Coding agents and save it in your tool’s skills folder. The prompts and the skill cite each thread by its uier link in pull requests and commits. The skill also shows your agent how to watch a project for new feedback, replies and status changes from a looping or scheduled task.
You stay in control
A connection works as your account, within the access you gave it.
Never more than you can do
Your current permissions are checked on every request. If you lose access to a project or a private thread, so does the connection.
Only what you choose
All your workspaces and projects, including new ones, or only the ones you pick. Consent is shown every time a tool connects.
Feedback is data, not instructions
The connector marks feedback as untrusted content and tells your agent to treat what reviewers wrote as text to read, not commands to follow.
Revoke at any time
Account → Coding agents lists your connections. Revoking one ends its access on the next request. It cannot take back what the tool already received.
Rewrite suggestions
When a reviewer wants different wording on the page, AI can suggest it. Try it here: select any sentence on this page.
- Select text on the page and choose Rewrite.
- Add an instruction if you like, such as shorter or friendlier, and choose Suggest.
- Admins can set the workspace’s tone, terms and style. Each project can add to or replace those instructions.
- Pick one of up to three suggestions. It goes into the text box, where you can edit it before you post the suggestion.
Anthropic’s Claude writes the suggestions from the selected text and your instruction, which are not stored; only what you post is saved. Suggestions work on selections of up to 1,000 characters and have hourly and daily limits. You can always write your own.
Try it
The MCP connector works with your projects. Set it up once and your agent can reach every project you allow.
- Sign in and create a project for your site.
- In Account, open Coding agents and set up your coding tool.
- Ask your agent to triage the open feedback, or to fix one thread.
Questions
More about uier in the main questions.
Which coding tools can connect?
Claude Code signs in through your browser. Codex and Cursor use an access token from Account → Coding agents. Other tools are added there once their connection has been checked. The MCP guide has the details.
Does uier send my feedback to an AI model?
Not through the MCP connector: your coding tool reads the feedback and handles it under its own terms, including what it sends to its model. Rewrite suggestions send the selected text, your instruction and any workspace or project style guidance to Anthropic. Thread titles send each thread’s first message to Anthropic, whose Claude Haiku writes a short title for lists, unless an admin turns them off. The privacy policy has the details.
Can an agent change things without asking?
A connection can do only what you allowed when it connected, and it is read-only unless you allow more. The prompts and the skill ask for your approval before every write, but the approval happens in your coding tool, so keep its own approval settings on.
Can an agent confirm a fix?
No. With Change status it can request or cancel a fix check after you approve that action. Only the reviewer confirms a fix; Fix verification explains how.
Does reading feedback mark it as read?
No. Threads your agent reads stay unread for you in the app and in the toolbar.
What does it cost?
Nothing. Every feature available today is free and will stay free forever. If we introduce pricing, it will apply only to new features users want.