Gemini Notebook for DeepSeek Harness
Gemini Notebook (NotebookLM) MCP plugin for DeepSeek Harness
Use Google NotebookLM / Gemini Notebook from the DeepSeek Harness (DSH) agent
53 tools for notebooks, sources, grounded Q&A, web research, audio podcasts, video, slides, quizzes and more.
English | 简体中文
dsh-gemini-notebook-mcp is a DeepSeek Harness plugin bundle (a "dsh plugin") that connects the
Gemini Notebook / NotebookLM MCP server
(notebooklm-mcp-cli) to the DSH agent. Once it's installed, you can ask DSH things like
"create a notebook about quantum computing, add these three URLs and generate a podcast"
and the agent does it through Gemini Notebook (formerly Google NotebookLM).
It contains configuration only, with no code of its own. It uses DSH's built-in MCP client, @deepseek-ai/dsh-mcp-client, to start
the notebooklm-mcp server over stdio. The server's tools then appear in DSH as
mcp__gemini-notebook__<tool>.
Contents
What is DeepSeek Harness?
DeepSeek Harness (DSH) is DeepSeek's
AI agent harness. Its motto is "Everything is a Plugin." It runs as a Desktop app, a
Web GUI (dsh web), a headless CLI, an ACP server or an SDK. Every capability,
including tools, UI panels, models, skills and MCP connections, is a
Cordis plugin. Plugins are composed into profiles
from bundles: npm-style packages that declare a cordis.patch.yml.
This repository is one of those bundles. It adds Gemini Notebook as an MCP tool provider to your
DSH profile.
Features
- 📚 Notebooks: list, create, rename, describe and delete Gemini Notebook notebooks.
- 🔗 Sources: add URLs, YouTube videos, pasted text, Google Drive docs and local files (PDF, DOCX, audio, images and more); read raw source text; sync stale Drive sources.
- 💬 Grounded Q&A: ask questions answered only from your sources, with citations. Queries can also run across several notebooks. Chats stay in the NotebookLM web history.
- 🔎 Research: fast or deep web research and Google Drive search, with import of the sources it finds.
- 🎙️ Studio: generate Audio Overviews (podcasts), video overviews, slide decks, infographics, reports, quizzes, flashcards, mind maps, data tables and interactive reports, then download them.
- 🗂️ Organization: notes, source labels, collections, tags, aliases, batch operations and multi-step pipelines.
- 🤝 Sharing: public links and collaborator invites.
- 👤 Accounts: multiple Google profiles, usage quota checks and NotebookLM Enterprise.
Requirements
Tested with DeepSeek Harness Desktop 0.2.1 on macOS and notebooklm-mcp-cli 0.15.4.
Installation
1. Install and sign in to the MCP server
uv tool install notebooklm-mcp-cli # installs nlm + notebooklm-mcp into ~/.local/bin
nlm login # opens a browser; sign in to Google once
nlm login --check # should print "Authentication valid!"
2. Install the DSH plugin bundle
Option A: ask the DSH agent, in a session where the agent has the plugin_manager tool (for example Creator mode):
Install the DSH bundle from github:EdgarsKalnins/dsh-gemini-notebook-mcp
Option B: Plugins page. Open Plugins in the DSH sidebar, choose install and enter
github:EdgarsKalnins/dsh-gemini-notebook-mcp.
Option C: from a local clone, which lets you edit the config before installing:
git clone https://github.com/EdgarsKalnins/dsh-gemini-notebook-mcp.git
Then ask the DSH agent: "Install the bundle at /absolute/path/to/dsh-gemini-notebook-mcp".
With hot reload (the default), the tools appear right away and no restart is needed.
3. Verify
Ask the agent: "Call gemini-notebook server_info and list my notebooks."
server_info should report auth_status: configured and 53 visible tools.
Usage examples
Talk to the DSH agent in plain language:
- "List my Gemini Notebook notebooks."
- "Create a notebook called AI Strategy Research and add https://example.com/article as a source."
- "Do deep research on cloud marketplace trends and import the cited sources."
- "What are the key findings in my Product Research notebook? Cite the sources."
- "Generate a deep-dive audio podcast from this notebook and download it when it's ready."
- "Make a 10-question hard quiz and a study guide from my lecture notes notebook."
- "Create a slide deck, then revise slide 3 to be more concise."
- "Which Google Drive sources in this notebook are out of date? Sync them."
Available tools
The tools are exposed as mcp__gemini-notebook__<name>.
| Group | Tools |
|---|
| Notebooks | notebook_list, notebook_get, notebook_describe, notebook_create, notebook_rename, notebook_delete |
| Sources | source_add, source_list_drive, source_describe, source_get_content, source_rename, source_sync_drive, source_delete |
| Chat & query | notebook_query, notebook_query_start, notebook_query_status, chat_configure, chat_list, chat_get, chat_export, cross_notebook_query |
| Studio | studio_create, studio_status, studio_revise, studio_delete, report, download_artifact, download_all_artifacts, export_artifact |
| Research | research_start, research_status, research_import |
| Notes | note, chat_save_to_note |
| Organization | label, tag, collection_list, collection_create, collection_edit, collection_set_emoji, collection_delete |
| Automation | batch, pipeline |
| Sharing | notebook_share_status, notebook_share_public, notebook_share_invite, notebook_share_batch |
| Account | profile, alias, refresh_auth, save_auth_tokens, usage_get, server_info |
Destructive tools such as notebook_delete, source_delete, studio_delete and sharing do nothing unless confirm=True is passed. Studio generation also asks for confirmation.
Configuration
All settings are in cordis.patch.yml:
- insert:
- id: gemini-notebook-mcp
name: '@deepseek-ai/dsh-mcp-client'
config:
serverName: gemini-notebook
transport: stdio
command: !!js '`${process.env.HOME}/.local/bin/notebooklm-mcp`'
args: ['--transport', 'stdio']
env:
NOTEBOOKLM_HL: en
NOTEBOOKLM_QUERY_TIMEOUT: '150'
toolCallTimeoutMs: 180000
failOnStartupError: true
Common changes:
| Goal | Change |
|---|
| Reduce context usage (53 tool schemas go with every request) | Add NOTEBOOKLM_DISABLED_GROUPS: studio,sharing,research under env |
| Hide or show single tools | NOTEBOOKLM_DISABLED_TOOLS / NOTEBOOKLM_ENABLED_TOOLS |
Run without installing (via uvx) | command: uvx, args: ['--from', 'notebooklm-mcp-cli', 'notebooklm-mcp'] |
| Binary in a different location | Set command to the output of which notebooklm-mcp |
| Download folder for artifacts | NOTEBOOKLM_DOWNLOAD_DIR: /path/to/dir |
| Language of the interface and generated artifacts | NOTEBOOKLM_HL: de (any BCP-47 tag, for example es-419) |
| NotebookLM Enterprise | NOTEBOOKLM_BASE_URL, NOTEBOOKLM_PROJECT_ID, NOTEBOOKLM_LOCATION |
Tool groups: notebooks_read, notebooks_manage, sources_read, sources_manage, chat,
query_multi, organization, automation, notes, auth, server, sharing, research, studio.
DSH scrubs secret-looking variables (*TOKEN*, *KEY*, …) from the server's environment. The
Google login lives in ~/.notebooklm-mcp-cli, and is encrypted there if you use nlm auth storage set protected.
Troubleshooting
| Symptom | Fix |
|---|
| Plugin fails to activate | Check that ~/.local/bin/notebooklm-mcp --help works, or fix command |
auth_status: stale or auth errors | Run nlm login, then call refresh_auth or reload the plugin |
| Queries time out on large notebooks | Raise NOTEBOOKLM_QUERY_TIMEOUT and toolCallTimeoutMs, or use notebook_query_start + notebook_query_status |
| Rate-limit / quota errors | Ask the agent to call usage_get; NotebookLM meters a rolling ~5 h window plus a weekly cap |
| Old server version | uv tool upgrade notebooklm-mcp-cli, then reload the plugin |
FAQ
How do I use NotebookLM in DeepSeek Harness?
Install notebooklm-mcp-cli, run nlm login, then install this bundle in DSH. The agent gets the full set of Gemini Notebook tools.
Is this an official Google or DeepSeek project?
No. It's a community DSH plugin. The MCP server comes from jacob-bd/gemini-notebook-mcp-cli and uses undocumented NotebookLM APIs.
Does it work with Claude Code, Cursor or Codex?
This bundle is for DeepSeek Harness only. For other agents, use upstream's nlm setup wizard.
Can I generate a NotebookLM podcast (Audio Overview) from DSH?
Yes. Ask the agent to create an audio artifact. It polls studio_status and downloads the result with download_artifact.
Does it support multiple Google accounts?
Yes. Run nlm login --profile work, then have the agent call profile(action="switch", name="work").
How it works
DeepSeek Harness agent
└─ @deepseek-ai/dsh-mcp-client (row "gemini-notebook-mcp", from this bundle's cordis.patch.yml)
└─ stdio ─► notebooklm-mcp (Python, notebooklm-mcp-cli)
└─ HTTPS ─► Gemini Notebook / NotebookLM (your Google session)
Repository layout:
| File | Purpose |
|---|
package.json | Bundle manifest (dsh.bundle.patch) |
cordis.patch.yml | Inserts the MCP client row |
locale/*.json | Title and description shown in DSH Plugin Manager |
icon.svg | Plugin icon |
Credits and disclaimer
NotebookLM and Gemini are trademarks of Google LLC. This project is not affiliated with Google or DeepSeek. The
upstream server uses undocumented APIs that may change, and use may not match Google's Terms of Service. Use it at your own risk.
License
MIT
