The Readwise MCP server lets Claude search across every highlight you’ve ever saved. Books, Kindle clippings, articles in Reader, podcast snippets. It’s the closest thing to a “search my reading history” tool that exists.
If you’ve been using Readwise for any length of time, you’ve probably saved thousands of highlights. The native Readwise UI lets you browse them. The MCP server lets Claude reason across them.
Why use it
The promise of Readwise has always been “your highlights, surfaced over time”. The daily review email is great. But the moment you want to actually find something, you’re scrolling through tags and search results.
Claude with the MCP server flips this. “Find every highlight I saved about pricing strategy.” “What did the books I read last year say about hiring?” Single prompt. The server queries Readwise. Claude synthesises across whatever comes back.
What it actually does
The first-party hosted server at mcp2.readwise.io/mcp covers both halves of the account, running hybrid vector and full-text search over your highlights and your Reader documents. It also writes, so Claude can create highlights, add tags, bulk edit document metadata and move documents. Authorization is OAuth in the browser.
The community servers wrap the same public API and stay read-only. They list books, filter highlights by book or tag or date, fetch your daily review, search Reader documents and pull article highlights, authenticating with a token from your Readwise account.
Practical patterns:
- “Pull every highlight I saved this month and group them by theme.”
- “What did Atomic Habits say about identity-based habits? Use my highlights, not the original book.”
- “Search my Reader articles for anything mentioning AI agents.”
Gotchas
The Readwise API has rate limits. Heavy queries (especially fetching all highlights for a large library) can take a while or hit the limit. The servers handle this gracefully but Claude may report partial results.
Highlight metadata varies. Books from Kindle have rich metadata. Articles from Reader have URL and author. Podcast snippets are sparse. Don’t expect perfectly uniform structured data.
For knowledge workflows, pair the Readwise server with ContextBolt. Readwise covers the books and articles you’ve read deeply. ContextBolt covers the social posts and threads you’ve bookmarked across X, Reddit, and LinkedIn. Together they give Claude a near-complete view of what you’ve consumed.
If you take notes in Notion or Obsidian, you can also build promotion workflows: “Find my best highlights from this month and add them to my Notion ‘Quote Library’ database.” Claude reads from Readwise, writes to Notion, in one prompt.