web-research-toolkit
v0.1.0
Published
Agent installer, prompts, and templates for web-research-mcp
Maintainers
Readme
Web Research Toolkit
Pre-built AI agents and skills for web research, compatible with Claude Code and OpenCode.
One command. Fully configured. Start researching.
Table of Contents
Quick Start
npx web-research-toolkit install opencodeor
npx web-research-toolkit install claude-codeThat's it. The installer configures the MCP server, installs agents, and sets up skills — no separate setup needed.
What It Does
Installs pre-built AI agents and skills into your project for web research:
- Agents — Autonomous subagents with defined workflows (quick search and deep research)
- Skills — On-demand expertise that agents load when needed (search methodology, research workflow)
- MCP server — Automatically configured to provide search and content extraction tools
Everything is installed locally in your project directory. One command, zero configuration.
What Gets Installed
| File | Claude Code | OpenCode |
|------|:-----------:|:--------:|
| MCP server config | .mcp.json | opencode.json |
| Plugin manifest | .claude-plugin/plugin.json | — |
| Web Searcher agent | agents/web-searcher.md | .opencode/agents/web-searcher.md |
| Deep Researcher agent | agents/deep-researcher.md | .opencode/agents/deep-researcher.md |
| Web Search skill | skills/web-search/SKILL.md | .opencode/skills/web-search/SKILL.md |
| Deep Research skill | skills/deep-research/SKILL.md | .opencode/skills/deep-research/SKILL.md |
Agents
web-searcher
Quick web search agent. Runs 1-3 targeted queries and returns concise answers with source URLs.
Good for:
- Factual lookups ("What's the latest version of React?")
- Finding documentation or API references
- Quick answers with sources
How it works: Formulates focused search queries, calls multi_search, and returns URLs, snippets, or direct answers. Fast and lightweight.
deep-researcher
Comprehensive multi-source research agent. Expands a research question into 10 diverse queries, searches all in parallel, fetches every result page, and synthesizes a detailed report.
Good for:
- Complex technical questions ("How do different ORMs handle connection pooling?")
- Comparative analysis ("Pros and cons of server-side rendering frameworks")
- Topics requiring multiple sources and synthesis
How it works: Generates diverse search queries, runs them all via multi_search, fetches full page content with fetch_pages, then synthesizes findings into a structured report with citations and confidence levels.
Skills
Skills are methodology guides that agents load on demand.
| Skill | What It Covers | |-------|---------------| | web-search | Search methodology — when to search vs. fetch, query optimization, result handling | | deep-research | Multi-phase research workflow — query expansion, source evaluation, synthesis |
Skills are loaded automatically when an agent determines they're relevant to the current task.
Architecture
Prompts are maintained once and work across platforms:
- Canonical prompts — Pure instructions, no platform-specific frontmatter
- Platform templates — Frontmatter and metadata specific to each platform (Claude Code, OpenCode)
- Installer — Assembles frontmatter + prompt body at install time, replacing
{{placeholders}}with metadata
Adding support for a new platform only requires new templates — the prompt content stays shared.
The MCP Server
This toolkit uses web-research-mcp under the hood. The installer configures it automatically — you don't need to set it up separately.
If you want just the MCP server without agents, see the web-research-mcp package.
