Research Tools

Tools for researching external sources from within Claude Code. Currently ships one skill:

• fetch-docs — pull a URL’s raw content into a local file so Claude reads the authoritative source instead of WebFetch’s summary. Works on any http/https URL, with optional HTML→markdown conversion.

Why

Claude Code’s built-in WebFetch runs a small-model pass that summarizes and filters the page through the user’s prompt. That pass routinely drops specifics — exact flag names, enum values, edge-case prose — the kind of detail you need when researching API shapes or CLI flags. fetch-docs bypasses that step: curl into /tmp/, return the path, Read it directly.

Installation

Run these in a terminal (not inside a Claude Code session):

claude plugin marketplace add jtsternberg/claude-plugins
claude plugin install research-tools@jtsternberg

Prerequisites

• curl — required for every call. Ships with macOS and almost every Linux distro.
• node / npx — only needed when you pass --md on an HTML source. Markdown-native URLs skip the conversion pipeline entirely. Both readability-cli and turndown-cli run through npx -y on first use, so there’s nothing to npm install.

# macOS
brew install node      # if npx isn't already on PATH

# Linux
# Use your distro's package manager or https://nodejs.org

Usage

Once installed, invoke the skill with /research-tools:fetch-docs <url> or let Claude trigger it when you ask for the raw page. Behind the scenes it runs the bundled scripts/fetch-docs.sh:

# Raw HTML (default)
bash plugins/research-tools/skills/fetch-docs/scripts/fetch-docs.sh "https://example.com/docs"

# HTML source → markdown
bash plugins/research-tools/skills/fetch-docs/scripts/fetch-docs.sh "https://example.com/docs" --md

# Markdown-native URL (auto-detected, no conversion run)
bash plugins/research-tools/skills/fetch-docs/scripts/fetch-docs.sh "https://code.claude.com/docs/en/skills.md"

# Custom slug + short TTL
bash plugins/research-tools/skills/fetch-docs/scripts/fetch-docs.sh "https://example.com/api" --slug=my-api --ttl=3600

The script prints the cached file path on stdout. See the SKILL.md for the full workflow, cache semantics, and troubleshooting.

Structure

plugins/research-tools/
├── .claude-plugin/plugin.json
├── README.md
└── skills/
    └── fetch-docs/
        ├── SKILL.md
        └── scripts/
            └── fetch-docs.sh

Roadmap

Other research-infrastructure skills that may live in this plugin over time:

• Multi-URL batch fetch
• Sitemap discovery and cached-docs library browser
• RSS/Atom readers
• Domain-scoped search helpers
• An auto-suggest hook that proposes fetch-docs when WebFetch is used on a docs URL