Layerdock

Читайте отзывы о сайте Layerdock в своём агенте для программирования. К каждому пину прилагаются CSS-селектор, скриншот, URL страницы, а также трассировка консоли и сети, зафиксированная в момент обращения клиента. Исправьте ошибку, ответьте в ветке и отметьте её решённой, не покидая редактор. Работает с Claude Code, Cursor, Codex CLI, Windsurf, VS Code, Zed и Antigravity.

Документация

How to connect your coding agent to Layerdock

Layerdock ships an MCP server, so any coding agent that speaks the Model Context Protocol can list your docks, read the feedback on them, reply on a thread and resolve a pin. The point is the round trip: “a reviewer flagged this” to “it’s fixed and they’ve been told” without leaving the editor. Setup is one token, one config block, one restart.

What your agent does once it’s connected

  1. **It reads the feedback.**Every open pin arrives with the CSS selector, the screenshot and the page URL, plus the console and network trace captured when the pin was reported and the element’s exact design specs from Inspect (spacing, radius, typography, colours). That full context comes with MCP on both Pro and Agency; viewing the console and network logs inside the dashboard is the Agency-only part.
  2. It fixes the thing. Your agent already has the repo open. It has the selector, so it knows which element the reviewer meant, not just which page.
  3. It replies and resolves. The reply lands on the thread your reviewer is watching and the pin flips to resolved. Nobody leaves the editor to close the loop.

Step 1: create a token

Open Integrations → Coding agents and name a token after the machine you will use it from, so you know which one to revoke later. Layerdock stores only a hash of it, so the plaintext value is shown exactly once, at creation — copy it before you leave the page. A token acts as you: it can read every dock you can reach and reply to feedback on your behalf, under the same role rules that apply to you in the dashboard.

Step 2: add the config for your editor

Pick your editor below and you get the one block it wants, in the file it wants it in. Replace ldk_your_token_here with the token you just copied. If the file already exists, merge the block into the object that is already there rather than pasting over it: a config that no longer parses is the most common reason the server never appears.

.mcp.json Repository root

{
  "mcpServers": {
    "layerdock": {
      "command": "npx",
      "args": ["-y", "@layerdock/mcp"],
      "env": { "LAYERDOCK_TOKEN": "ldk_your_token_here" }
    }
  }
}

This file is normally committed. On a shared repo, export LAYERDOCK_TOKEN in your shell and put "${LAYERDOCK_TOKEN}" in the env value instead of the token itself, so your personal token never lands in git history.

Before you restart: warm the cache

Run this once, in a terminal, before your editor tries to connect:

npx -y @layerdock/mcp --version

npx downloads the package the first time it runs it, and on a slow connection that download can take longer than your editor allows an MCP server to start — 30 seconds in Claude Code. The result is a CONNECT_TIMEOUT on a server that is working perfectly. The command above puts it in the npx cache and exits; every launch after that is instant. If you have already hit the timeout, run it and restart your editor.

Any other MCP client

Nothing above is specific to those seven editors. Any MCP client works: give it the command npx -y @layerdock/mcp and the environment variable LAYERDOCK_TOKEN, in whichever file or settings screen it keeps MCP servers in. The package is published on npm as @layerdock/mcp and needs no global install; npx fetches it on first run.

Step 3: check it worked

  1. Restart your editor. MCP servers are read once, at startup. Quit the application and open it again; reloading the window is not always enough.
  2. Confirm it connected. In Claude Code, type /mcp and Layerdock should be listed as connected. In any other client, ask the agent to list the tools it has and look for the Layerdock ones.
  3. Give it something to do. Open a dock that has feedback on it and try the prompt below.

Read the open Layerdock feedback on this dock, fix what you can, then reply on each pin and resolve it.

If the server does not show up

**The client never lists the server.**Almost always the file, not the token. Check that it still parses after your edit (one stray comma breaks the whole config), that you merged the block into the existing object rather than pasting over it, and that you used the block for this client — VS Code and Zed both name the key differently.

It times out on the very first launch. Almost always a cold npx download outrunning the client’s start-up budget, not a problem with the server. Run npx -y @layerdock/mcp --version once in a terminal and restart your editor — the second launch reads from the npx cache and connects immediately.

It connects, but every call comes back unauthorized. The token is wrong or no longer valid. Confirm the prefix in your config matches a token listed as Active on the Coding agents page, and that you pasted the whole value. Tokens cannot be read back, so if you are unsure, revoke it and create a new one.

Related

For what the agent actually receives and why it matters, see AI bug fixing with MCP. This guide assumes you already have a dock with feedback on it. If not, start with collecting client feedback on a website, or see pricing.