Web Research

A web research toolkit for AI coding assistants — search, extract, and synthesize with pre-built agents.

No API keys. One command to install. Works with Claude Code and OpenCode.

---

Table of Contents

• Quick Start
• What's Included
• Agents & Skills
• MCP Tools
• Development
• Contributing
• License

---

Quick Start

Install agents and skills for your platform:

npx web-research-toolkit install claude-code

or

npx web-research-toolkit install opencode

That's it. The installer configures the MCP server, installs agents, and sets up skills — no separate setup needed.

Just want the MCP server?

If you only need the search and fetch tools (without agents), add the server to your MCP client config directly:

{
  "mcpServers": {
    "web-research": {
      "command": "npx",
      "args": ["web-research-mcp"]
    }
  }
}

---

What's Included

This monorepo ships two npm packages:

Package	Description
web-research-mcp	MCP server with two tools: multi_search (DuckDuckGo) and fetch_pages (parallel content extraction)
web-research-toolkit	Installs pre-built AI agents and skills into your project for Claude Code or OpenCode

---

Agents & Skills

The toolkit installs agents (autonomous subagents with defined workflows) and skills (on-demand expertise that agents load when needed).

Agents:

• web-searcher — Quick web search. Runs a few targeted queries and returns URLs, snippets, or direct answers.
• deep-researcher — Comprehensive multi-source research. Expands a question into 10 diverse queries, fetches every result, and synthesizes a detailed report with citations.

Skills:

• web-search — Search methodology and result handling
• deep-research — Multi-phase research workflow

See the web-research-toolkit README for full details on each agent and skill.

---

MCP Tools

The MCP server provides two tools for AI assistants:

• multi_search — Search DuckDuckGo with multiple queries in parallel. Returns deduplicated URLs, snippets, and titles.
• fetch_pages — Fetch and extract clean text from multiple web pages in parallel. Uses Mozilla Readability for content extraction.

No API keys required. Works with any MCP-compatible client.

See the web-research-mcp README for tool parameters and examples.

---

Development

npm install          # install all workspace deps
npm run build        # build both packages
npm test             # test both packages
npm run typecheck    # type check both packages

Monorepo structure:

packages/
├── mcp/       # web-research-mcp — MCP server
└── toolkit/   # web-research-toolkit — agent installer

---

Contributing

Contributions are welcome! Please open an issue or submit a pull request.

---

License

MIT © Nacho F. Lizaur