Skip to main content
Glama

@yocoolab/mcp-server

npm version npm downloads CI License Node Types OpenSSF Best Practices

MCP (Model Context Protocol) server that exposes Yocoolab feedback threads and design selections as tools for Claude Code and other MCP-compatible clients.

With this installed, your AI coding assistant can:

  • List and triage open design feedback threads on your repo

  • Pull rich context for a thread (selection, screenshot, conversation, files touched)

  • Reply to designers and mark threads as addressed

  • Open PRs that close out feedback threads

  • Inspect element context and selection history from the Yocoolab Chrome extension

  • Read companion chat messages, AI conversations, and Pendo product analytics

Most people no longer need this package. Yocoolab hosts the MCP server at https://app.yocoolab.com/mcp. Your agent signs in to Yocoolab in the browser the first time, so there is no token to paste:

claude mcp add --transport http --scope user yocoolab https://app.yocoolab.com/mcp

Then type /mcp in Claude Code, choose yocoolab, and Authenticate. Setup for Cursor, VS Code, the Claude app, and headless agents (agent token in a header) is at app.yocoolab.com/connect.

Always-on agent: to have one computer pick up every Send to Agent in your workspace by itself (it runs Claude Code), run this on that computer:

curl -fsSL https://app.yocoolab.com/agent/install.sh | sh

Switching from this package to the hosted server? Remove the old entry first (claude mcp remove yocoolab, repeat until none is found). Otherwise Claude may keep using the old one.

Use this package only if you need what runs on your own machine: the local bridge on :9800 for live selections from the Chrome extension.

Related MCP server: Claude Talk to Figma MCP

Install (local package)

One command:

npx -y @yocoolab/mcp-server@3 setup

The setup wizard auto-detects your installed AI agents (Claude Code, Cursor, Cline, Roo Code, Windsurf) and writes the correct MCP config for each. Restart your agent and the yocoolab server appears with all tools available.

init is kept as an alias for setup for backwards compatibility with v1.0.x install instructions.

If you'd rather configure manually, the equivalent ~/.mcp.json looks like:

{
  "mcpServers": {
    "yocoolab": {
      "command": "npx",
      "args": ["-y", "@yocoolab/mcp-server@3"],
      "env": {
        "YOCOOLAB_API_URL": "https://app.yocoolab.com",
        "YOCOOLAB_TOKEN": "<your-yocoolab-jwt>",
        "GITHUB_TOKEN": "<your-github-pat>",
        "YOCOOLAB_BRIDGE_PORT": "9800",
        "YOCOOLAB_BRIDGE_WORKSPACE": "/absolute/path/to/your/workspace"
      }
    }
  }
}

The @3 version pin keeps you on the v3 major line — you'll receive bug fixes and new features automatically, but a future v4 with breaking changes won't break your setup. v3.0.0 removed the activity monitor and no longer writes Claude Code hooks into ~/.claude/settings.json; see CHANGELOG.md if you are upgrading from v2.

Requirements

  • Node.js 20 or newer. We test on Node 20 and 22 in CI. We support whichever Node.js versions are currently in Active LTS or Maintenance LTS status, and drop versions within 30 days of their EOL. Node 18 was dropped in v2.0.0 (EOL April 2025).

  • A Yocoolab account and JWT token — get yours from the Yocoolab Chrome extension settings, or via your account at app.yocoolab.com.

  • A GitHub personal access token with repo scope, if you want to use the PR-creation tools (create_pr_for_thread). The token is auto-detected if you have the GitHub CLI installed and authenticated (gh auth login). No GITHUB_TOKEN env var needed in that case. Otherwise, create one at github.com/settings/tokens/new?scopes=repo.

Configuration

Env var

Required

Default

Description

YOCOOLAB_TOKEN

no

—

Your Yocoolab JWT (from the Chrome extension). When unset, thread feedback tools are disabled but bridge / companion / selection tools still work.

YOCOOLAB_API_URL

no

https://app.yocoolab.com

Yocoolab API base URL

GITHUB_TOKEN

only for PR tools

—

GitHub PAT with repo scope. Auto-detected — if gh CLI is installed and authenticated (gh auth token), no env var is needed. Placeholder values like <your GitHub PAT> are detected and safely ignored.

YOCOOLAB_BRIDGE_PORT

no

9800

Local port for the HTTP bridge to the Chrome extension

YOCOOLAB_BRIDGE_WORKSPACE

no

process.cwd()

Absolute path to your project workspace, used to resolve file references in selections

YOCOOLAB_AGENT_NAME

no

Claude Code

Display name shown in the Chrome extension's agent picker

YOCOOLAB_PROJECT_PR_LINKS

no

1

Set to 0 to restore the pre-2.5 behaviour: mark_thread_addressed without PR/preview link fields, GitHub-only PR guidance in the prompt

YOCOOLAB_AGENT_TYPE

no

claude-code

Agent type identifier (claude-code, roo, cline, cursor, windsurf, or custom)

CLI

yocoolab-mcp           Run the MCP server (used by your agent via .mcp.json)
yocoolab-mcp setup     Interactive setup — auto-detects agents and writes their configs
yocoolab-mcp init      Alias for `setup` (backwards compatible with v1.0.x)
yocoolab-mcp --help    Show this help

The mcp-server command is a synonym for yocoolab-mcp. Either works.

Tools

The server exposes tools across several categories:

  • Threads — list_open_threads, get_thread_context, add_thread_message, mark_thread_addressed, create_pr_for_thread

    • On a link-only project (no GitHub App connection) open the PR with your own git/gh, then call mark_thread_addressed with pr_url (and pr_branch, preview_url when you have them). The PR fields land on the thread and the preview link goes into the review message; get_thread_context prints the project's repo links so you know which repository it points at. create_pr_for_thread remains for GitHub-connected owner/name workspaces.

  • Selection / Bridge — get_latest_selection, get_selection_history, get_element_context, find_source_for_selection, ai_analyze_page

  • Companion — get_companion_messages, reply_to_companion

  • AI — get_ai_conversations

  • Deployment — get_deployment_preview

  • Pendo (optional) — pendo_list_guides, pendo_page_analytics, pendo_feature_usage, pendo_track_event

For full tool descriptions and parameters, your MCP client will list them after the server starts.

Troubleshooting

yocoolab-mcp: command not found — make sure you're on v1.0.1 or newer. Run npx -y @yocoolab/mcp-server@latest setup to get the current release.

[yocoolab] Warning: YOCOOLAB_TOKEN not set — thread feedback tools are disabled without a token, but bridge / companion / selection tools still work. To enable everything, run yocoolab-mcp setup to (re)generate the config with your JWT.

Tools don't appear in your agent after install — restart your agent completely (quit & reopen). MCP servers load at startup.

Port 9800 is already in use — another instance of the MCP server is running, or another app has the port. Set YOCOOLAB_BRIDGE_PORT to a different value (e.g. 9801) in your .mcp.json.

Verbose diagnostic logs — set DEBUG=yocoolab:* in your env block. All diagnostic output goes to stderr (so it doesn't interfere with the MCP stdio protocol on stdout).

PR creation says "GitHub token not configured" — the create_pr_for_thread tool needs a GitHub token. Two ways to fix:

  1. Auto-detect (easiest): install the GitHub CLI and run gh auth login. The server picks up the token automatically.

  2. Manual: create a personal access token with repo scope and add GITHUB_TOKEN: "ghp_..." to your MCP config's env block. If your config has a placeholder value like <your GitHub PAT>, the server detects it and shows a helpful message instead of crashing with a 401.

Support

Development

git clone https://github.com/Yocoolab/mcp-server.git
cd mcp-server
npm install
npm run build       # compile TypeScript to dist/
npm test            # run the vitest suite
npm run dev         # tsc --watch

See CONTRIBUTING.md for the full contributor workflow.

Security & supply chain

  • Released with npm provenance — every published version is cryptographically signed by GitHub Actions OIDC, traceable back to the exact commit and workflow run.

  • Each release ships with a CycloneDX SBOM attached to the GitHub Release.

  • We run npm audit signatures and CodeQL static analysis in CI on every PR.

  • See SECURITY.md for vulnerability reporting.

License

Apache 2.0 — © 2026 Yocoolab

Available Tools

10 tools
find_source_for_selectionA

Searches the workspace for source files that likely define or render the selected UI element. Returns ranked candidates with file paths, line ranges, confidence scores, and match reasons. Can work with live Chrome selections or stored thread element_context.

Inputs: correlationId, thread_id.

get_activity_summaryA

Get a high-level summary of Claude Code activity: total events, active sessions, tool usage breakdown, files touched, and error count.

Inputs: no named parameters.

get_companion_messagesA

Check for new messages from the AI Companion panel in the Chrome extension. Returns pending messages and clears the queue. Each message includes the user text, page context, and optional screenshot/element attachments.

Inputs: no named parameters.

get_dashboard_urlA

Get the URL for the real-time activity monitoring dashboard. Open in a browser to see live Claude Code activity, tool usage charts, and session timelines.

Inputs: no named parameters.

get_element_contextA

Returns a comprehensive summary of a UI element, including its DOM context, computed styles, and ranked source code candidates. Works with live Chrome selections or stored thread element_context. This is the primary tool to use when the user wants to modify a UI element.

Inputs: correlationId, thread_id, includePrompt.

get_files_touchedA

List all files that have been read, written, or edited across Claude Code sessions. Shows file paths with session counts.

Inputs: no named parameters.

get_latest_selectionA

Returns the most recent element selection from the Chrome extension. Includes element selector, tag, text content, classes, attributes, computed styles, and page context.

Inputs: no named parameters.

get_recent_eventsA

Get the most recent Claude Code activity events (tool calls, prompts, errors). Shows timestamps, tool names, and key details.

Inputs: limit.

get_selection_historyA

Returns the last N element selections from the Chrome extension, newest first. Useful for comparing multiple elements or reviewing recent selections.

Inputs: limit.

reply_to_companionA

Send a reply back to the AI Companion panel in the Chrome extension. The reply will appear as an assistant message in the companion chat UI. Supports Markdown formatting.

Inputs: message (required).

Tool Schema Changelog

Recent tool additions, removals, and schema changes observed during successful MCP inspections.

Loading tool schema changelog…

    TDQS

    A4.1/5.0

    Scored across 10 tools

    Disambiguation5/5

    Each tool has a clearly distinct purpose: selection handling, source finding, element context, companion messaging, activity monitoring, and dashboard URL. Even similar tools like get_latest_selection and get_selection_history differ by scope (single vs. history). No overlap that would cause confusion.

    Naming Consistency4/5

    All tools follow a verb_noun pattern with underscores (e.g., get_activity_summary, find_source_for_selection). While most use 'get', two use 'find' and 'reply', which is a minor deviation from a uniform verb prefix. Overall pattern is consistent and predictable.

    Tool Count5/5

    With 10 tools, the server is well-scoped for its purpose of combining UI element inspection and Claude Code activity monitoring. Each tool serves a clear role without unnecessary bloat or deficiency.

    Completeness4/5

    The tool surface covers the key workflows: selecting elements, getting context, finding source code, companion chat, and activity monitoring. Minor gaps exist, such as no tool to clear selection history or reset companion state, but these are not critical for the core use case.

    Maintenance

    ActivitySlowing
    ResponsivenessNo recent community issues

    Related MCP Connectors

    Related MCP Servers

    • A
      license
      Not graded
      quality
      D
      maintenance
      Enables AI assistants like Claude and Cursor to interact directly with Figma to create, modify, and analyze design elements in real-time. It provides a comprehensive suite of tools for document inspection, styling, component management, and automated layout via a bidirectional WebSocket connection.
      MIT
    • A
      license
      Not graded
      quality
      D
      maintenance
      Enables AI assistants to inspect, modify, export, and validate design documents with 47 tools covering design, code generation, branding, and print/mockup workflows.
      2
      MIT
    • A
      license
      A
      quality
      C
      maintenance
      Enables visual browser feedback collection directly into Claude Code. Users can point at elements in their browser and send annotated feedback that Claude can act on immediately.
      12
      1
      MIT