Skip to main content
skill CX Codex

agentation — Claude Code skill: visual feedback toolbar for Next.js

Adds a visual feedback toolbar to a Next.js project and, on request, wires it to your coding agent over MCP.

Install into Claude Code · Skills CLI (Vercel) All methods and platforms ↓
npx skills add benjitaylor/agentation --skill agentation -a claude-code
Open on GitHub

What agentation does

The skill walks the agent through setup. First it reads the project's instructions, detects the package manager, and checks for agentation and an existing <Agentation> mount. If the component is already there, the agent inspects its endpoint and current MCP setup instead of adding a second mount. The package goes in with whichever manager owns the lockfile.

Then comes the connection mode. Live agent feedback needs the same HTTP URL the agent uses; the default local server is http://localhost:4747, and a custom port or proxy URL is kept as is. On a remote project you cannot assume the browser and the agent share a machine. Manual copy/paste mode omits endpoint, and feedback stays in that browser.

In App Router the component renders in the root layout's body after children; in Pages Router, after Component in pages/_app. The package ships its own client boundary, so the root layout stays server-rendered. For MCP the launch command carries the server subcommand: npx -y agentation-mcp server. React 18 or newer is required, and installing the toolbar neither installs nor connects an MCP server. A healthy doctor run is not proof of sync: npx agentation-mcp doctor reports service state, while only a test annotation read back through agentation_get_all_pending confirms the live path.

What it helps with

  • mount a visual feedback toolbar in a Next.js app
  • connect browser annotations to a coding agent over MCP
  • avoid starting a second HTTP server on a busy port
  • tell real sync apart from a merely healthy server
  • keep the toolbar out of production builds

Example prompts for the agent

  • Add the Agentation toolbar to our Next.js project and connect it to the agent over MCP.
  • Verify that browser annotations reach the agent and tell me which part is still unverified.
  • Set up Agentation in manual copy mode, without a server.

How to install agentation

1Skills CLI (Vercel)

npx skills add benjitaylor/agentation --skill agentation -a claude-code

2Manually: put the files in place

~/.claude/skills/agentation/SKILL.md

Copy the whole skill folder; for one project use .claude/skills/agentation/

Step by step, with commands for each platform: How to install skills in Claude Code: five ways

Questions about agentation

How do I install agentation in Claude Code?

Run npx skills add benjitaylor/agentation --skill agentation -a claude-code. The skill lands in ~/.claude/skills/agentation/SKILL.md.

What does agentation do?

It adds a visual feedback toolbar to a Next.js project and, when asked, connects it to your coding agent through an MCP server.

Do I need the MCP server for the toolbar to work?

No. Manual copy mode runs without a server, but then feedback stays in the browser and never reaches the agent.

Source text

SKILL.md · MIT
---
name: agentation
description: Add Agentation visual feedback toolbar to a Next.js project
---

# Agentation Setup

Set up the toolbar and, when requested, connect it to the user's coding agent.
A visible toolbar and a healthy MCP server do not prove that browser feedback
reaches the agent. Verify that final step before calling setup complete.

## Steps

1. **Inspect the project**
   - Read the project's instructions and detect its package manager.
   - Look for `agentation` and an existing `<Agentation>` mount before installing.
   - If already mounted, inspect its `endpoint` and current MCP setup. Do not
     exit just because the component exists, and do not add a duplicate mount.
   - Install `agentation` if missing, using the existing lockfile's package manager.

2. **Choose the connection mode from the request**
   - For MCP or live agent feedback, use the same HTTP server URL as the agent.
     The default local server is `http://localhost:4747`; preserve an existing
     custom port or proxy URL. Never assume the browser and agent use the same
     machine when the project is remote.
   - For manual copy/paste only, omit `endpoint` and explain that feedback remains
     in this browser. Do not silently enable a server connection.
   - If the intended mode is unclear, ask one brief question while inspecting
     the existing project. Keep working on independent setup checks.

3. **Add or update the component**

   For a Next.js App Router project, render in the root layout's body after
   children. For Pages Router, render after `Component` in `pages/_app`.
   Preserve any existing callbacks and options.

   ```tsx
   import { Agentation } from "agentation";

   // Use the actual server URL selected above for MCP sync:
   {process.env.NODE_ENV === "development" && (
     <Agentation endpoint="http://localhost:4747" />
   )}
   ```

   The package provides its own client boundary. Keep the application's root
   layout server-rendered unless the project has another reason to change it.

4. **Configure the agent when MCP was requested**
   - Use the project's existing MCP client and configuration conventions.
   - The launch command must include the `server` subcommand:
     `npx -y agentation-mcp server`.
   - For supported agents, `npx add-mcp "npx -y agentation-mcp server"` is one
     setup option. For Claude Code, use
     `claude mcp add agentation -- npx -y agentation-mcp server` or the init wizard.
   - Apply the same custom port to the server command and component endpoint.
   - Restart or reconnect the agent when its configuration requires it. Do not
     create a second HTTP server on a port already owned by the working server.

5. **Verify the complete connection**
   - Run `npx agentation-mcp doctor` for service diagnostics. For a custom URL,
     use `doctor --http-url <url>` when supported by the installed version.
   - Open the app and create one clearly identified test annotation through the
     toolbar. Read it with `agentation_get_all_pending` or session-scoped tools.
   - Check the exact test comment and page, not just a nonzero count or health
     response. Leave unrelated annotations alone.
   - If annotations are empty, check the component endpoint, browser network
     failures/CORS, and whether the agent and browser point to the same server.
   - If browser or MCP tools are unavailable, say exactly which part remains
     unverified and give the user this final check. Do not claim live sync works.

## Notes

- The development guard prevents the toolbar from appearing in production.
- Agentation requires React 18 or newer.
- Installing the toolbar does not install, configure or connect an MCP server.
- Manual copy mode is supported and does not need a running server.

© benjitaylor · MIT · benjitaylor/agentation

Similar skills

All skills for Agent setup: memory, context, orchestration →

From the same set2

Project sponsors

Companies that keep this analytics open
Gold sponsor

Impressions, clicks and code copies — open statistics. We keep the numbers public for the same reason NeBlask itself is public. Become a sponsor