OAuth 2.1, not API keys
Every platform signs in directly through your browser using OAuth with CIMD or PKCE. Never paste passwords, sync tokens, or model keys into chat.
PaperEnjoyer MCP · v0.1.1
Connect your AI assistant to PaperEnjoyer. Add papers, read your library, start a close reading, and save annotations. Changes sync to your devices.
https://api.paperenjoyer.com/mcp“Add this paper to my PaperEnjoyer: https://arxiv.org/abs/1706.03762”
Imported into your cloud library. Automatic parsing has started—your desktop and mobile apps will pull the parsed paper on their next sync.
“Find the paper about attention in my PaperEnjoyer and check Section 3.2.”
Found Attention Is All You Need. In Section 3.2.1, dividing dot products by √dk prevents softmax gradients from saturating at large dimensions.
“Start a close reading of it in PaperEnjoyer and save a note on scaling.”
PaperCore close reading has started in the cloud, and your note is anchored to Section 3.2.1 so you can jump back to the exact sentence.
“Move the duplicate draft paper to trash.”
Moving "Attention Draft (2017)" to the Trash (recoverable; permanent deletion is never exposed). Please confirm if you want me to run action_execute.
How it works
Every platform signs in directly through your browser using OAuth with CIMD or PKCE. Never paste passwords, sync tokens, or model keys into chat.
Tools read and write your synced cloud library even when your desktop app is closed. Enabled desktop and mobile clients pull updates automatically.
Reading and importing proceed smoothly. Moving items to trash, cancelling tasks, or editing someone else’s notes preview exact impact and wait for your confirmation.
Cloud MCP toolset
Inspect folder trees, tags, and counts, or search by title, author, abstract, DOI, and arXiv ID across your synced cloud collection.
Paginate original PDFs and close-reading editions paragraph-by-paragraph with verified snapshots and UTF-16 character offsets.
Send PDF, arXiv, OpenReview, or DOI URLs. PaperEnjoyer fetches and parses them automatically via MinerU and PaperCore.
Supports unencrypted PDFs up to 200 MiB and 200 pages—via the plugin upload helper or a private browser upload link on this page.
Trigger cloud close reading when you ask for PaperEnjoyer, track task status, and pick up finished insights on any synced device.
Create folders, move papers, refine metadata, and save paragraph-anchored annotations directly onto originals or close-reading notes.
01 · Choose your platform
Pick your AI platform below for step-by-step setup, copyable commands, and the full in-site tutorial and troubleshooting guide.
Connect directly in ChatGPT Web Plugins first; the downloadable ZIP is an optional companion workflow Skill.
Requirements:ChatGPT account & workspace with custom MCP enabled
Quick start
Why connect first, download second
From v0.1.1 onward, ChatGPT uses a custom remote MCP connection created directly under your own account—no shared publisher App ID or public store listing is needed.
The optional paperenjoyer-chatgpt-0.1.1.zip contains only the Skill instructions and upload helper script. It intentionally excludes MCP connection config or a Connect button in plugin settings.
If you previously installed paperenjoyer-codex-0.1.0.zip in ChatGPT, create and verify your new custom MCP connection first, remove the old combined package, then import the new workflow ZIP.
Common fixes
No “Add custom MCP server” button: check whether your ChatGPT plan or workspace administrator restricts custom MCP connectors; importing the workflow ZIP will not bypass this.
No PaperEnjoyer tools appear in chat: confirm you installed the connection you created and selected it in the new conversation.
Connection revoked or expired: re-enable the connection in the Connections section below, then re-authorize that MCP connection inside ChatGPT.
Agent Plugins 1.0 package with native MCP config, Codex compatibility manifest, and research workflow skills.
Requirements:Codex with plugin marketplace or local ZIP support
Quick start
Detailed guide & configuration
codex plugin marketplace add watericetangcw/PaperEnjoyer-PluginsPackage structure & PDF uploads
The Codex package ships plugin.json, mcp.json, skills/, and a .codex-plugin/ compatibility manifest. Do not copy SKILL.md alone without mcp.json.
ChatGPT users should follow the ChatGPT tab above rather than using the Codex package.
When importing a local PDF that the host cannot send safely through the helper script, ask Codex for the returned web upload link and upload the PDF in your browser below.
Common fixes
Tools missing after install: make sure you installed the full plugin from the marketplace rather than copying only the SKILL.md file.
Authentication expired: reconnect PaperEnjoyer inside Codex MCP settings; if revoked below, re-enable it here first.
Install directly via marketplace in Claude Code, or add a custom remote connector in Claude Web / Desktop / Cowork.
Requirements:Claude Code CLI, or Claude Web / Desktop / Cowork with Connectors
Quick start
Detailed guide & configuration
/plugin marketplace add watericetangcw/PaperEnjoyer-Plugins
/plugin install paperenjoyer@paperenjoyerclaude --plugin-dir /absolute/path/to/paperenjoyerClaude Code vs. Claude Web / Desktop
Claude Code loads .claude-plugin/plugin.json, .mcp.json, and skills together from paperenjoyer-claude-0.1.1.zip or the marketplace.
Claude Web / Desktop / Cowork uses two separate steps: (1) adding the Remote Connector with “Use Claude’s published identity” (never pick “No sign in” or paste tokens), and (2) optionally uploading paperenjoyer-claude-skill-0.1.1.zip in Custom Skills.
Wait for an import task to finish parsing before starting PaperCore close reading on that paper.
Common fixes
Connector asks for credentials or fails to sign in: verify OAuth client is set to “Use Claude’s published identity” rather than “No sign in”.
Skill upload rejected in Claude Web: ensure you uploaded paperenjoyer-claude-skill-0.1.1.zip rather than paperenjoyer-claude-0.1.1.zip.
Installs straight from the plugin Git repository or .tgz archive using a local stdio OAuth bridge.
Requirements:DeepSeekHarness 0.2.0-rc.2+ and Node 22.19+ in PATH
Quick start
Detailed guide & configuration
https://github.com/watericetangcw/PaperEnjoyer-Plugins.gitdsh plugin --profile desktop add https://github.com/watericetangcw/PaperEnjoyer-Plugins.gitPinning versions & remote SSH sessions
Enabling the plugin never triggers paid PaperCore reading automatically.
For reproducible Git updates in DSH, pin a full 40-character commit SHA from the release notes (for example ...PaperEnjoyer-Plugins.git#<commit-sha>) rather than a placeholder.
If DSH runs on a remote SSH host, the OAuth callback targets 127.0.0.1 on that host—run DSH locally or forward the loopback callback port via SSH; never paste authorization codes into chat.
Common fixes
Bridge fails to start: confirm Node 22.19+ is available on the PATH used by DSH.
Need to re-authenticate: call paperenjoyer_connect again; credentials live in your OS keychain when supported.
Registers PaperEnjoyer MCP and Skill transforms via @opencode/plugin, plus ready-to-paste OpenCode 1 config.
Requirements:OpenCode 2 (or OpenCode 1 with remote MCP OAuth support)
Quick start
Detailed guide & configuration
npm install --ignore-scripts{
"$schema": "https://opencode.ai/config.json",
"plugins": ["/absolute/path/to/paperenjoyer"]
}{
"$schema": "https://opencode.ai/config.json",
"mcp": {
"paperenjoyer": {
"type": "remote",
"url": "https://api.paperenjoyer.com/mcp",
"oauth": {}
}
}
}OpenCode 2 vs. OpenCode 1
OpenCode 2 uses ctx.mcp.transform and ctx.skill.transform; existing entries with the same name take precedence.
Still on OpenCode 1? Merge the legacy mcp block above into your existing config without overwriting your other plugins/servers, and copy skills/paperenjoyer into your skills folder. Do not load the OpenCode 2 plugin entry in v1.
Common fixes
Plugin fails to load: verify you ran npm install --ignore-scripts inside the extracted directory and used an absolute path in opencode.json.
Already defined a custom paperenjoyer MCP entry: check that its URL still points to https://api.paperenjoyer.com/mcp.
Dedicated Pi package exposing native registerMcpServer integration and PaperEnjoyer research skills.
Requirements:Pi version with native MCP (registerMcpServer)
Quick start
Detailed guide & configuration
pi install /absolute/path/to/packageWhy use the dedicated .tgz package
The dedicated .tgz avoids asking Pi’s Git installer to pick a subdirectory out of the multi-platform plugin repository—do not install the repository root directly in Pi.
No third-party MCP adapter is needed when Pi supports native registerMcpServer.
If you also declared a same-named MCP server in a config file, Pi prefers the file entry—ensure its URL matches https://api.paperenjoyer.com/mcp.
Common fixes
Error mentioning registerMcpServer is missing: upgrade Pi to a release with native MCP support before reloading.
Native Qoder manifest, mcp.json, and skills using an automatic pinned OAuth stdio bridge.
Requirements:Node 22.19+ with npx reachable by Qoder
Quick start
Bridge authentication & tool refresh
Qoder’s mcp.json fetches the pinned bridge release over npx and connects via stdio—no long-lived bearer token setup is required.
Once browser login succeeds, the tool list updates automatically; if your Qoder version does not refresh immediately, reconnect the MCP server once.
System keychain credentials persist sessions across restarts when available.
Common fixes
Tools do not appear right after login: toggle/reconnect the PaperEnjoyer MCP entry in Qoder once.
Bridge startup error: verify node -v is 22.19+ and npx works in a standard shell.
Install straight from the GitHub plugin marketplace or via local ZIP import.
Requirements:Node 22.19+ with npx reachable by ZCode
Quick start
Detailed guide & configuration
watericetangcw/PaperEnjoyer-PluginsWhere tools appear & safety prompts
Tools appear under the Plugin MCP Servers group once connected.
Do not paste passwords or tokens into chat—authentication happens purely through the browser OAuth PKCE flow.
Deleting papers or cancelling tasks always triggers action_prepare first so you can review exact impacts before confirming.
Common fixes
First launch takes a moment: npx downloads the pinned bridge package once from GitHub Releases.
Need to switch or revoke accounts: disconnect below on this page, re-enable if needed, and re-run paperenjoyer_connect.
Prompt cookbook
Click any prompt to copy it. Generic requests to read a paper stay with your AI assistant unless you name PaperEnjoyer or continue a paper already there.
“Add this paper to my PaperEnjoyer.”
“Find the paper about attention in my PaperEnjoyer.”
“Start a close reading of it in PaperEnjoyer.”
“List the 5 most recently added papers in my PaperEnjoyer.”
Browsing your library, reading passages, checking quotas, and importing requested papers run smoothly. Duplicate PDFs reuse existing records without double-charging.
Imports parse automatically (small credit usage when PaperCore applies). Close reading runs in the cloud using PaperCore credits—no external model keys stored here.
Moving papers or folders to trash, cancelling tasks, or changing notes from another source require action_prepare first, with a 5-minute single-use preview awaiting your consent.
Importing includes parsing and may use a small amount of credits. Close reading uses PaperCore credits. Deleting papers, cancelling tasks, and changing someone else’s annotations require confirmation.
Manage platforms authorized to access your PaperEnjoyer cloud library.
Disconnecting immediately blocks new requests from that platform while submitted background tasks continue. Re-enabling generates a fresh connection ID—re-authorize in your AI client to get a new token.
Choose a PDF, up to 200 MiB and 200 pages. Parsing starts automatically.
MCP Q & A
No. MCP tools read and write directly to your synced PaperEnjoyer cloud library. Parsing and PaperCore close reading continue in the cloud, and desktop or mobile apps with sync enabled pull updates when opened.
Cloud MCP can only see papers, folders, and notes that have been synced to your PaperEnjoyer account. Enable cloud sync in your desktop settings for local papers you want accessible from AI assistants.
No permanent delete tool exists in MCP—deletions only move items to Trash. Furthermore, trashing papers or folders, cancelling tasks, and modifying notes from other sources require action_prepare to show exact targets before action_execute.
Ask your assistant to prepare an upload for PaperEnjoyer. It will generate a private link to this page (?upload=...) where you can sign in and upload an unencrypted PDF up to 200 MiB and 200 pages.