This project provides AI coding agents with "Lazyweb skills" to perform design research and ideation using real-world web and application screenshots. It addresses the limitation of AI models designing based solely on training data averages by integrating a hosted Lazyweb MCP server, enabling agents to search for, download, and analyze current design examples. This empowers agents to produce evidence-based designs, conduct competitive analysis, and generate actionable improvement ideas, benefiting designers and developers seeking to ground their work in current industry practices.

How It Works

The core of the system is the Lazyweb MCP server, which serves as a backend for retrieving real application and web screenshots. AI agents utilize specific skills to interact with this server, performing searches based on descriptions, categories, companies, or platforms. The retrieved screenshots are downloaded locally and organized into structured reports (e.g., report.md) with inline image previews, facilitating detailed analysis of design patterns, competitors, and potential improvements. This approach allows agents to move beyond generic outputs by grounding their design decisions in current, tangible visual data.

Quick Start & Requirements

This repository is packaged as both a Codex and a Claude Code plugin.

• Codex Plugin Setup:
  • Install by placing plugin source in plugins/lazyweb/, skills in ~/plugins/lazyweb/skills/, MCP config in ~/plugins/lazyweb/.mcp.json, and marketplace entry in ~/.agents/plugins/marketplace.json.
  • Set LAZYWEB_MCP_TOKEN environment variable or store the generated token at ~/.lazyweb/lazyweb_mcp_token. The plugin also checks the legacy ~/.codex/lazyweb_mcp_token path.
• Claude Code Plugin Setup:
  • Add the marketplace: claude plugin marketplace add https://github.com/aboul3ata/lazyweb-skill
  • Install the plugin: claude plugin install lazyweb@lazyweb
  • Claude Code skills are namespaced as /lazyweb:<skill-name>.
• Token Generation: A free bearer token is required for MCP reference tools. Generate it by running:

  mkdir -p ~/.lazyweb
  curl -sS -X POST https://www.lazyweb.com/api/mcp/install-token \
    -H "content-type: application/json" \
    -d '{}' | node -e "let s='';process.stdin.on('data',d=>s+=d);process.stdin.on('end',()=>require('fs').writeFileSync(process.env.HOME+'/.lazyweb/lazyweb_mcp_token', JSON.parse(s).token))"
  

• Prerequisites: curl, node, and a generated Lazyweb MCP token stored locally.

Highlighted Details

• Skills: Includes /lazyweb-design-research (deep analysis), /lazyweb-quick-references (fast visual grouping), /lazyweb-design-improve (actionable critique), /lazyweb-design-brainstorm (cross-category inspiration), and tools for managing inspiration sources.
• MCP Tooling: Utilizes lazyweb_search (with filters for category, company, platform), lazyweb_compare_image, and lazyweb_find_similar. These map to canonical MCP tools like search_screenshots and vision_screenshots.
• Output Structure: Generates .lazyweb/{skill}/{topic}-{date}/report.md with inline images and a references/ directory for downloaded screenshots.

Maintenance & Community

No specific details regarding maintainers, sponsorships, or community channels (like Discord/Slack) are provided in the README.

Licensing & Compatibility

The project is licensed under the MIT License. The generated bearer token is for authorizing Lazyweb MCP reference tools only and does not grant access to purchases, paid services, private user data, or destructive actions. The token should be kept out of public git repositories.

Limitations & Caveats

The Lazyweb MCP token's scope is limited to reference tools, excluding any paid features, private data access, or modification capabilities. The primary utility is as a plugin for AI coding agents, suggesting its integration is within that specific ecosystem.