6sense
Ask an AI agent about 6sense account insights, buying signals and keyword trends.
Search every highlight you have saved and triage your Reader inbox.
Opens Agent Studio, where connecting is one click. The connector URL below works in any MCP client.
Paste it into any MCP client. Setup docs
The Readwise MCP server connects your reading library to Claude, ChatGPT, Codex and any MCP-compatible agent. It searches every highlight you have saved from books and articles, triages your Reader inbox, tags and archives documents, and creates new highlights in conversation. Sign-in is OAuth, and it is one of the few connectors here offering separate read and write scopes.
Verified connector
Listed by Anthropic as a partner connector in its Connectors Directory.
Connection checked by Agentman on .
Anthropic states this reflects the level of review a connector received, not a security audit.
Tool names from Anthropic's directory listing. This server requires sign-in, so we could not read tool descriptions or parameter schemas.
readwise_* tools working against a library the sixteen reader_* tools have nothing to read — and all five tag tools are Reader tools, so they go with it.readOnlyHint or destructiveHint on any tool.entry.permissions is absent from Anthropic's snapshot for this server. The single cheapest read/write cross-check is unavailable, and we state that rather than substituting a quieter inference.https://mcp.readwise.io/sse answers anonymously, exposes two tools, reaches Readwise highlights only, and publishes a descriptor with no scopes. Readwise's own legacy guide still prints that URL. Check which endpoint your client holds.reader_bulk_edit_document_metadata at 50 documents at a time, and its skills repository puts the same 50-per-call cap on reader_move_documents and states the two tools share a rate limit of 20 calls per minute, advising you to batch IDs into fewer calls. A conversational "archive everything older than a month" can reach both ceilings.429 carries a Retry-After header with the seconds to wait. Readwise does not state how MCP tool calls map onto those counters, so treat the figures as platform-wide rather than connector-specific.Both, through one connection. Readwise states the integration indexes your Readwise highlights and your Reader documents and makes all of them available to connected AI apps. That is why the tool names split into two prefixes: six readwise ones covering saved highlights, and sixteen reader ones covering the read-later library.
It replaces it. Readwise states the previous Readwise-only server is being deprecated in favour of this one, which covers both products. Readwise tells anyone who configured the old server to update their AI app to the new URL. We found the old host still answering on 2026-08-18, so the switch is on you to make.
Yes for tags and moves. Readwise states every action the MCP takes is reversible the same way you would undo it in Reader, by removing the tag, moving the document back or restoring it from archive, and that you can ask the assistant to undo what it just did. Deletion is a separate question.
Continuously, in the background. Readwise states the index updates automatically as you add or change content, and that after the initial full index later syncs only process new or changed material so they finish quickly. A document saved moments ago may still be mid-index when an assistant searches for it.
Yes, with no stated ceiling. Readwise states you can use the same MCP server URL with as many compatible apps as you like. Each app runs its own OAuth authorisation, so every connection is a separate grant that you approve and can revoke on its own without disturbing the others.
No. Readwise states there is no free plan after the free trial, which runs 30 days. The connector fronts a paid subscription, so budget for it before wiring an agent to it. Readwise does document discounts for students, teachers, non-profits, military, first responders and people in countries with depressed currencies.
Because Readwise stores a soft discard rather than a hard delete. Readwise explains it keeps the record so a re-sync from Kindle or another source does not re-import the highlight as new, and says outright that a hard delete would frustrate users wanting to recover an accidental deletion. Readwise directs anyone needing true erasure to delete at the original source.
Yes, from Trash, with one catch that matters here. Readwise states deleted documents stay in Trash until you recover them or empty it, and that restoring returns the document to your library along with its highlights and notes. Readwise also states the Trash is device-specific, so a document must be restored from the same device that deleted it.
initialize against https://mcp2.readwise.io/mcp — HTTP 401 with WWW-Authenticate: Bearer resource_metadata="https://mcp2.readwise.io/.well-known/oauth-protected-resource/mcp", our own observation, 2026-08-18. This challenge is what establishes the OAuth posture. A plain GET to the same URL also returns 401, so no unauthenticated tool catalogue is served. · retrieved 2026-08-18authorization_servers: ["https://readwise.io/o/"] and scopes_supported: ["openid", "read", "write"]. The path-less form on the same host returns 404, so only the RFC-insert form is served. · retrieved 2026-08-18issuer: https://readwise.io/o/, a registration_endpoint, authorization_code / implicit / refresh_token grants, code_challenge_methods_supported: ["S256"], and the same three scopes. The OpenID discovery document at also returns HTTP 200 and lists the same three scopes. · retrieved 2026-08-18POST https://mcp.readwise.io/mcp returns 404; GET https://mcp.readwise.io/sse returns HTTP 200 with content-type: text/event-stream and issues a session endpoint. An anonymous initialize over that session succeeds without authentication, reporting serverInfo.name: "Readwise MCP (v0.0.2)", version 1.9.4 and protocol 2025-03-26. Its tools/list returns exactly two tools, search and fetch; prompts/list and resources/list are declared in initialize capabilities and both return empty arrays. Its descriptor at returns HTTP 200 naming resource: "https://mcp.readwise.io" with no scopes_supported key. We ran only listing RPCs and never called a tool. · retrieved 2026-08-18mcp-remote bridge command, and six example prompts. · retrieved 2026-08-18docs.readwise.io/sitemap.xml; no sub-URL was guessed. · retrieved 2026-08-18Z / shake-to-undo feature. · retrieved 2026-08-18429 wait-time response. · retrieved 2026-08-18https://docs.readwise.io/llms.txt (HTTP 200, 473 KB, retrieved 2026-08-18). Source for the vendor still publishing https://mcp.readwise.io/sse as the ChatGPT MCP Server URL on the same page that announces the deprecation, and for the local @readwise/readwise-mcp stdio server taking an ACCESS_TOKEN. · retrieved 2026-08-18readwiseio/readwise-skills (default branch master, pushed 2026-06-19, retrieved 2026-08-18). Reached from a real link on https://readwise.io/mcp; the repository URL was not guessed. Source for the third independent tool list, the eleven published skills, the .mcp.json naming the same mcp2 endpoint, the 50- document batch ceiling, the 20-calls-per-minute shared rate limit, and the two-product description. reader_persona appears there as a generated file name, not a tool. · retrieved 2026-06-19robots.txt for readwise.io (HTTP 200, retrieved 2026-08-18). Allow: / with two narrow disallows, neither on a path used here, and no Content-Signal directive on any of the three axes, so fetching and synthesis are both unrestricted. readwise.io/llms.txt returns 404; readwise.io/sitemap.xml returns 200 and is how the pricing pages were located. · retrieved 2026-08-18robots.txt for docs.readwise.io (HTTP 200, retrieved 2026-08-18). Allow: /, no Content-Signal, and it names the sitemap. The host serves both an llms.txt and a sitemap.xml; the four /tools/* pages that carry this page's best material appear in the sitemap and not in llms.txt. · retrieved 2026-08-18permissions field. Anthropic's directory documentation states that verification "is not a security audit". · retrieved 2026-08-16Connect Readwise once and your agents call these tools on their own: on a schedule, in a workflow, with nobody at the keyboard.