This Will Transform Your Conversations with Claude
Based on Readwise's video on YouTube. If you like this content, support the original creators by watching, liking and subscribing to their content.
Import highlights into Readwise first, using Connect and Sync for sources like Kindle Pocket and Reader, and use the physical book import tool for paper books.
Briefing
Readwise’s new MCP server turns Claude into a personal “knowledge layer” over a user’s own reading—so answers can be generated from highlights stored in Readwise, not just from the open web. The practical payoff is faster retrieval of exact quotes and more reliable synthesis of ideas across years of Kindle, podcasts, YouTube, and even physical books, all inside Claude.
The workflow starts in Readwise: users must import their highlights from multiple sources. The dashboard offers “Connect and Sync” to bring in highlights from over 30 sources, including Kindle Pocket and Readwise’s own read-it-later app, Reader (available in a free trial and included with a full Readwise subscription). For people who highlight physical books, Readwise also provides a physical book import tool inside the Readwise mobile app. Once the highlights are in Readwise, Claude can be configured to access them.
Installation requires Claude’s desktop environment and a small amount of technical setup. Claude currently supports MCP in its desktop app, so users need the Claude desktop app installed on the relevant device. They also need a way to edit JSON files; the tutorial recommends Visual Studio Code for both macOS and PC. In Claude, users go to Settings → Developer → “Edit Configuration,” then open the Clog Desktop JSON file in Visual Studio Code. The Readwise MCP code is copied from the Readwise MCP page and pasted into the Claude JSON configuration.
A key step is authorization: the configuration includes an “Access Token” field. Users copy an access token from the Readwise MCP page and paste it into the JSON file between quotation marks. After saving, the recommended move is to restart Claude so the MCP integration takes effect.
Once installed, Claude can query the user’s highlight database using semantic (fuzzy) search. Instead of needing exact wording, a user can describe what a quote was about—e.g., trying to recall a highlight about being yourself in relationships—and Claude searches the highlight vectors in Readwise to surface the most relevant matches. The example shows Claude returning the exact quote the user remembered, along with additional similar highlights from another book the user had highlighted.
Because Claude’s responses can mix direct quotations with its own commentary, the tutorial suggests a formatting prompt: ask Claude to rewrite the response so direct quotes are clearly separated (bolded/italicized) while Claude’s own commentary remains plain text. That makes it easier to lift quotes for writing or journaling.
The MCP also supports synthesis. For an academic literature review use case, the user prompts Claude to identify main themes scholars have found in a short story, explicitly cueing Claude to rely on Readwise highlights rather than the web. Claude then produces a structured, scan-friendly synthesis with bolded headers and italicized excerpts tied to the underlying highlights. The same approach can be applied to work problem-solving or creative/personal questions.
Finally, the tutorial notes that Readwise MCP is installable in Claude and Cursor, with a rumor that ChatGPT may support MCP soon—at which point a follow-up tutorial is expected.
Cornell Notes
Readwise’s MCP server lets Claude answer using a user’s own highlights stored in Readwise, including Kindle, podcasts, YouTube, and physical books. After importing highlights into Readwise, users install the Readwise MCP code into Claude’s desktop JSON configuration and add an Access Token for permission. Once connected, Claude can retrieve exact quotes via semantic search even when the user only remembers the idea, not the wording. It can also synthesize themes across many highlights for tasks like academic literature reviews, producing structured outputs that reference the underlying excerpts. This matters because it makes Claude’s responses more grounded in personal reading history and easier to reuse for writing and research.
What setup steps are required before Claude can use Readwise highlights?
How does semantic (fuzzy) search work for finding a remembered quote?
Why might Claude’s output be confusing, and how can formatting fix it?
How can the MCP be used for synthesis rather than quote retrieval?
What kinds of tasks beyond academics fit this workflow?
Review Questions
- What is the role of the Access Token in the Readwise MCP installation for Claude?
- How would you prompt Claude to retrieve a quote when you remember only the topic, not the exact wording?
- What formatting instruction would you give Claude to clearly separate direct quotes from its own commentary?
Key Points
- 1
Import highlights into Readwise first, using Connect and Sync for sources like Kindle Pocket and Reader, and use the physical book import tool for paper books.
- 2
Claude’s MCP support is desktop-only, so the integration requires the Claude desktop app on the target device.
- 3
Edit the Claude configuration by opening the Clog Desktop JSON file in Visual Studio Code and pasting the Readwise MCP code into it.
- 4
Add the Access Token from the Readwise MCP page into the JSON configuration to authorize Claude to access Readwise highlights.
- 5
Use semantic search by describing the idea behind a quote; Claude can retrieve the exact highlight even without exact wording.
- 6
Ask for formatting rules (e.g., emphasized direct quotes vs plain-text commentary) to make extracted quotes easy to scan and reuse.
- 7
Prompt for synthesis when you need themes across many highlights, with instructions that steer Claude toward Readwise rather than the web.