MCP for design feedback
Send design feedback to your coding agent.
Connect Claude Code, Codex or Cursor over MCP. The agent pulls every open item with the selector and the screenshot, fixes it in your repository, and reports back on each one.
Included in every plan. 14-day free trial.
One token, one command.
Then ask your agent.
01Create a token
Under Settings, AI access, create a token. It is shown once and stored hashed. One command registers the server in Claude Code or Codex.
02Ask the agent
Inside your repository: pull the open feedback from my layernote project and fix it. The agent opens each item with its screenshot and selector.
03Report back on every item
The agent replies in the thread with what it changed and moves the card on the board, or resolves it. Everything it does carries your name.
Built for client work.
Your code stays on your machine.
Reaches what you reach
A token belongs to one person and one workspace. It reaches only the projects you reach there, and everything it does carries your name.
Named, hashed, revocable
Give every token a name you recognise. Tokens are stored hashed and can only be revoked, under Settings, AI access, which stops them at once.
Any agent that speaks MCP
layernote documents Claude Code, Codex CLI and Cursor; other clients configure the same npx command.
Your code stays yours
The MCP server runs on your machine and only talks to layernote. It never sends your code anywhere.
Admins see every token
Workspace owners and admins see every token and connected app of their team, and can revoke any of them.
Included in every plan
AI access is part of Solo, Studio and Team, for as long as the workspace’s plan is active.
In practice
One prompt. Every open item handled.
You ask once. The agent lists the open feedback, fixes each item in your code and closes it in layernote with what it changed.
- list_feedback finds the open comments on a page.
- get_feedback opens each one with its selector and screenshot.
- add_feedback_comment and resolve_feedback close it with what changed.
get_feedback
What the agent gets per item
Not a paraphrase of the comment: the element, the thread and the screenshot. Enough to find the code and know what to change.
- The page URL and the CSS selector of the element
- The comment with its full thread
- The element screenshot, as an image the agent can look at
Tools
The tools the server exposes
The tools map onto the same statuses your team sees on the board, so an agent’s work shows up like anyone else’s: replies in the thread, cards moving across columns, your name on everything it does.
- Static files projects get file tools too: list_files, get_file, compare_versions and add_comments.
- update_feedback assigns a comment, or sets its priority or due date.
- Won’t fix needs a reason, for false positives and out-of-scope requests.
The Model Context Protocol is the open standard coding agents such as Claude Code, Codex and Cursor use to talk to external tools. layernote exposes a project’s feedback through it, so the agent reads comments and screenshots directly instead of you copying them into a prompt.
Any agent that speaks MCP. layernote documents Claude Code, Codex CLI and Cursor; other clients configure the same npx command.
No. The MCP server only supplies data and actions to the agent you already run and pay for.
Yes. The agent’s work shows up like anyone else’s: replies in the thread, cards moving across the board, and your name on everything it does, because the token is yours.
Yes. AI access is part of Solo, Studio and Team, for as long as the workspace’s plan is active.
Because feedback shouldn’t be the hard part.
One link for your clients to review, one list for your team to resolve.