Install in one line
Claude Code
claude mcp add toolkb -- npx -y github:jack-reson8/toolkb-mcp
Codex (~/.codex/config.toml)
[mcp_servers.toolkb] command = "npx" args = ["-y", "github:jack-reson8/toolkb-mcp"]
Cursor, Claude Desktop, Windsurf (mcp.json)
{ "mcpServers": { "toolkb": { "command": "npx", "args": ["-y", "github:jack-reson8/toolkb-mcp"] } } }
Needs Node.js 18+. Runs locally, no API key, no account. Behind a proxy? HTTPS_PROXY is picked up automatically.
What your assistant gets
# How to Do Great Work - Source: https://www.paulgraham.com/greatwork.html - Summary: If you collected lists of techniques for doing great work… --- The first step is to decide what to work on…
Things to ask
- “Read these three docs pages and write a migration guide from v1 to v2.”
- “Turn this blog post into a checklist I can follow.”
- “Compare the pricing pages of these two products in a table.”
Also included
The same server gives your assistant YouTube transcripts, whole playlists and channels, web pages as Markdown and App Store reviews. There is also a terminal command: npx -y -p github:jack-reson8/toolkb-mcp toolkb yt <url>.
FAQ
Why not just let the assistant fetch the page?
Built-in fetchers often return raw HTML or truncated text. toolkb isolates the article body, keeps headings, lists, code blocks and links, and makes every link absolute.
Does it handle documentation sites?
Yes. When the article extractor finds too little text it falls back to the page's main content area, which suits docs and knowledge bases.
Can it read PDFs?
Not yet – PDF links are on the roadmap.
Is there a length limit?
Output is capped at 60,000 characters by default; set max_chars to change it.