Here is a workflow that did not exist a year ago: instead of copy-pasting articles into a chat window, you let your AI assistant reach into your reading list directly. Ask "what did I save about sleep training?" and get an answer drawn from your own library, with sources. This guide shows how to set it up with Keepfor and Claude in a few minutes.
What MCP is, in one paragraph
MCP, the Model Context Protocol, is an open standard that lets AI assistants use external tools. Think of it as USB for AI: instead of each assistant building custom integrations for every app, any app can expose tools once and any assistant can use them. Keepfor runs an MCP server with 47 tools covering its whole library: searching, reading, tagging, saving, and organizing. Claude Desktop, Claude Code, and Cursor all speak MCP, so they can all use it.
What you can actually do
Once connected, your assistant is not guessing. It is working with your real saved articles. Examples people use daily:
- "What did I save about urban gardening?" Claude searches your library by keyword and by meaning, then answers from your own articles.
- "Summarize the three longest unread articles I saved this month." It reads them and gives you the gist.
- "Tag everything I saved last week." It suggests tags based on content, and applies the ones you approve.
- "Save this thread for later." Paste a link in chat and it lands in your Keepfor library, tagged.
- "What should I read next?" It looks at what you have been saving and reading and finds related items already in your library.
- "Give me my weekly digest." A summary of what you saved and what is worth your attention.
Under the hood these map to real tools: search_library (hybrid keyword plus semantic search), read_item, suggest_tags, tag_item, save_url, save_note, pin_item, related_items, and library_digest, among others.
Setup
You need a Keepfor account. The hosted plans include a 14-day free trial with no card required; the open-source core can also be self-hosted for free.
Step 1: Create an access token. In the Keepfor web app, open Settings and create a Personal Access Token. This is what authenticates your assistant. Treat it like a password: it grants full access to your library.
Step 2: Add the server to your assistant. Keepfor publishes ready-made configuration snippets for Claude Desktop, Claude Code, and Cursor at keepfor.me, in the MCP section. The server itself lives at https://app.keepfor.me/api/mcp and speaks streamable HTTP. For Claude Desktop, paste the snippet into your MCP configuration file and restart the app.
Step 3: Approve the tools. The first time Claude wants to touch your library, your MCP client will ask you to approve the tools. Approve the read tools freely; be more deliberate with the write tools (save_url, tag_item, delete_item) until you trust the workflow.
That is the whole setup. Most people are asking their first question within five minutes.
Honest notes
- It costs quota, not just money. Questions you ask over your library draw on your plan's monthly AI quota: 50 ask-your-library queries a month on Personal, 500 on Pro. Heavy daily use belongs on Pro.
- Your assistant sees what you let it see. The token grants access to your library, so only connect assistants you trust, and use a dedicated token you can revoke from Settings at any time.
- It is not Claude-only. Anything that speaks MCP works: Cursor, Claude Code, and custom agents. The setup pattern is the same everywhere.
- Self-hosters get this too. The MCP server is part of the open-source core, so a self-hosted Keepfor exposes the same tools on your own infrastructure.
Why this matters beyond convenience
Read-it-later apps have always had the same failure mode: the archive grows and nobody ever reads it again, because finding anything requires remembering it exists. An assistant that can search, summarize, and resurface your own library turns the archive back into something you use. That is the whole idea behind Keepfor's "AI context vault" positioning, and MCP is the mechanism that makes it real.
Sources: MCP server details from the Keepfor open-source repository (47 tools, streamable HTTP endpoint); connection flow from keepfor.me/llms.txt; pricing and quotas from keepfor.me/pricing.