NotebookLM MCP Server Guide 2026: Connect NotebookLM to Claude

Table of Contents

1. What Is the NotebookLM MCP Server?
2. Why Connect NotebookLM to Claude
3. Which NotebookLM MCP Should You Use?
4. What the Tools Let an Agent Do
5. How to Set Up the NotebookLM MCP
6. Authentication and Persistent Sessions
7. NotebookLM MCP vs Context7 vs DeepWiki
8. Real Use Cases
9. Limitations
10. Verdict
11. Frequently Asked Questions

What Is the NotebookLM MCP Server?

NotebookLM is Google's research tool that turns a set of documents you upload — PDFs, docs, slides, web pages — into a grounded assistant that answers questions only from those sources, with citations. It is exceptional at staying anchored to your material instead of making things up. The NotebookLM MCP server exposes that capability over the Model Context Protocol, so an AI client like Claude, Cursor, or Codex can query your notebooks directly inside a conversation or coding session.

The pitch is simple and compelling: every answer your agent gets from NotebookLM is grounded in your uploaded knowledge base, with citations, which eliminates invented APIs and outdated information. Instead of pasting documents into a chat window over and over, you point the agent at your notebooks once and let it research, synthesize, and reference across your documents as part of its normal workflow.

Because Google does not publish an official server for this, the NotebookLM MCP ecosystem is community-built — and surprisingly rich, with several mature implementations to choose from.

Why Connect NotebookLM to Claude

NotebookLM and Claude have complementary strengths, and connecting them covers each one's blind spot.

NotebookLM is brilliant at grounded recall: ask it something and it answers strictly from your sources, with citations, refusing to wander. But it is a closed box — it cannot reach outside your uploads, and it cannot write code, run tools, or take action.

Claude is the opposite: it reasons, codes, and acts, and it can pull in outside context. What it lacks is your private corpus — the internal docs, the research PDFs, the project files sitting in your notebooks.

Bridge them with an MCP server and you get the best of both. Claude queries NotebookLM for citation-backed facts from your documents, then does something with them: writes the integration code, drafts the summary, combines your private knowledge with external information NotebookLM alone could never access. As one reviewer put it, you get to "take advantage of Claude's explanations and coding skills" while every factual claim stays grounded in your own sources. For research, documentation, and codebase work, that combination is genuinely powerful.

Which NotebookLM MCP Should You Use?

There is no single canonical package — "NotebookLM MCP" is a family of community projects. The ones worth knowing, all indexed on Skiln:

• PleasePrompto/notebooklm-mcp — a popular implementation focused on letting agents research documentation with grounded, citation-backed answers, with persistent auth and library management.
• alfredang/notebooklm-mcp — full programmatic access: create notebooks, manage sources, and generate content like podcasts, videos, slides, mind maps, quizzes, flashcards, and reports through natural language.
• mulyg/notebooklm-mcp and other Glama-listed builds — lighter servers focused on the core query loop.
• jacob-bd/notebooklm-mcp-cli — programmatic access via CLI, MCP server, and an agent skill, for people who want all three surfaces.
• julianoczkowski/notebooklm-mcp-2026 — a query-focused server that works across Claude Code, Claude Desktop, Cursor, and VS Code.

Some of these drive a real Chrome browser (via tools like Patchright) so the agent can read DOM-level citations and even generate audio overviews; others run as a lightweight local stdio subprocess that lists notebooks and asks questions with no HTTP server needed. Start with a well-starred, recently-updated option and check Skiln for current stars and maintenance before committing.

What the Tools Let an Agent Do

Capabilities vary by implementation, but the fuller servers expose on the order of 16 tools. The common building blocks:

• List notebooks — discover which notebooks exist in your account.
• Read source content — pull the text of a notebook's underlying sources into context.
• Ask a question — send a natural-language query to NotebookLM and get a grounded, citation-backed answer from that notebook's sources.
• Add sources — ingest new documents or URLs into a notebook.
• Generate content — in the richer servers, produce audio overviews, videos, mind maps, quizzes, flashcards, and reports from a notebook.

In practice the agent chains these: list notebooks, pick the relevant one, ask targeted questions, and optionally generate a study artifact. The lighter servers keep to the first three — list, read, ask — which is enough for most research and Q&A workflows.

How to Set Up the NotebookLM MCP

Exact steps depend on the implementation, but the shape is consistent. You generally need the Claude Desktop or Claude Code client and uv, a fast Python package manager.

1. Install uv if you do not have it, then install or fetch the server package per its README.
2. Authenticate with Google. Run the server's login command, which opens a browser for a one-time Google sign-in so the MCP can reach your NotebookLM account.
3. Add a config entry so your client knows the server exists. A typical Claude Desktop entry looks like:

{   "mcpServers": {     "notebooklm": {       "command": "uvx",       "args": [""]     }   } }

1. Restart the client, confirm the NotebookLM tools appear in the MCP list, and test with: "List my NotebookLM notebooks, then ask the research notebook how the method handles edge cases."

The Skiln Config Generator can fold NotebookLM into a multi-server config alongside your other tools.

Authentication and Persistent Sessions

This is the part that makes NotebookLM MCP servers different from a typical API integration. Because NotebookLM has no public API for this, the servers authenticate by logging into your Google account through a browser — a one-time OAuth-style sign-in. Better implementations then persist that session so you are not logging in on every run, storing the auth state locally.

Some servers go further and drive a real Chrome browser under the hood. That is how they read notebook content and DOM-level citations, and how they trigger generation features like audio overviews. The upside is access to capabilities the web UI exposes; the downside is fragility — if Google changes the NotebookLM interface, a browser-driving server can break until it is updated.

Practical advice: pick an actively maintained server, expect the occasional update when Google ships UI changes, and keep your persistent session file secure since it grants access to your Google account.

NotebookLM MCP vs Context7 vs DeepWiki

Three ways to give an agent grounded knowledge, compared:

These are complementary. Use the NotebookLM MCP to ground answers in your private documents and research. Use the Context7 MCP for up-to-date public library documentation while coding (see our Context7 review). Use the DeepWiki MCP to understand public GitHub codebases. A research-heavy agent might run all three, choosing the source that fits the question.

Real Use Cases

• Research synthesis. Upload a stack of papers or reports to a notebook, then have Claude query it for grounded findings and write a synthesis with citations.
• Codebase documentation. Point a notebook at your internal docs and let an agent answer "how does our auth flow work?" with references, then write code against the real spec.
• Grounded writing. Draft content where every factual claim is anchored to a source you control, reducing hallucination risk.
• Study artifacts. With the richer servers, generate flashcards, mind maps, or an audio overview from your notebook for learning.
• Mixed research. Combine NotebookLM's grounded recall from your sources with Claude's ability to fetch outside context, getting answers NotebookLM alone could not produce.

For broader research stacks, pair it with the best research and academic MCP servers and the best memory and knowledge graph MCP servers.

Limitations

• No official server. The whole ecosystem is community-built, so quality and maintenance vary between implementations.
• Browser-based auth is fragile. Servers that drive Chrome or scrape the NotebookLM UI can break when Google changes the interface, until they are patched.
• Depends on NotebookLM's limits. You inherit NotebookLM's source caps and rate limits; heavy use may want NotebookLM Pro.
• Session security. A persistent Google session file is sensitive — protect it like a credential.
• Setup friction. uv, a browser login, and a config file are more steps than a remote, no-auth server like DeepWiki.

Verdict

The NotebookLM MCP server turns Google's best-in-class grounded research tool into something your agent can use directly, and the result is genuinely one of the better research setups available in 2026. Citation-backed answers from your own documents, combined with Claude's reasoning and coding, is a powerful loop that cuts hallucinations and eliminates endless copy-paste. The richer implementations even let the agent generate audio overviews and study artifacts.

The caveats are real — no official server, browser-based auth that can break on Google's UI changes, and more setup than a remote server — so pick an actively maintained implementation and expect occasional updates. But for anyone doing serious research, documentation, or knowledge work, connecting NotebookLM to your agent is well worth the setup. Add it, then layer in research MCP servers for a complete grounded-knowledge stack.

Frequently Asked Questions

What is the NotebookLM MCP server?

The NotebookLM MCP server is a Model Context Protocol server that connects an AI client like Claude, Cursor, or Codex to Google's NotebookLM, so the agent can list your notebooks, read sources, and ask grounded, citation-backed questions about your uploaded documents. Because NotebookLM answers only from your sources, the integration gives your agent accurate, current knowledge with citations instead of hallucinated answers.

Why connect NotebookLM to Claude instead of using NotebookLM alone?

NotebookLM is excellent at grounded answers from your sources but cannot reach beyond them or write code. Claude can reason, code, and pull in outside context but lacks your private knowledge base. Connecting them lets Claude query NotebookLM for citation-backed facts from your documents, then act on those facts — writing code, drafting docs, or combining them with external information NotebookLM alone cannot access.

How do I install a NotebookLM MCP server?

Most implementations need the Claude Desktop (or Claude Code) client and uv, a Python package manager. You install the server, run a one-time browser login so it can authenticate with your Google account, then add a config entry pointing your client at the server. After restarting the client, the NotebookLM tools appear and the agent can list notebooks and ask questions. Setup specifics vary by implementation, so follow the README for the one you choose.

Does the NotebookLM MCP require a Google login?

Yes. Because NotebookLM is a Google product without a public API for this, the MCP servers authenticate through a one-time browser login to your Google account. Many implementations persist that session so you do not log in every time. Some drive a real Chrome browser under the hood to read notebooks and citations at the DOM level.

Is there an official NotebookLM MCP server?

Google does not ship an official NotebookLM MCP server, so the ecosystem is community-built. Popular options include PleasePrompto/notebooklm-mcp, alfredang/notebooklm-mcp, jacob-bd/notebooklm-mcp-cli, and julianoczkowski/notebooklm-mcp-2026, each with different capabilities such as querying, source management, or generating audio overviews. Skiln indexes the leading implementations so you can compare stars and maintenance.

How many tools does the NotebookLM MCP expose?

It depends on the implementation. Fuller servers expose on the order of 16 tools covering notebook management, querying sources, adding sources, and generating content like audio overviews, videos, mind maps, and flashcards. Lighter servers focus on the core loop: list notebooks, read source content, and ask the NotebookLM AI questions about your sources via a local stdio subprocess.

Is the NotebookLM MCP free?

The MCP servers themselves are open source and free. You need a Google account with NotebookLM access, which has a free tier, and NotebookLM Pro unlocks higher limits. You also pay for your AI client usage. There is no separate license fee for the community MCP servers, though they depend on NotebookLM's web interface and can break if Google changes it.

Where can I find other research and knowledge MCP servers?

Skiln indexes research, documentation, and knowledge MCP servers across every major registry. Read our roundups of the best research and academic MCP servers and the best memory and knowledge graph MCP servers, or browse the category directly at /mcps to compare implementations.

Last updated: June 28, 2026 · Skiln tracks new MCP servers daily across 13 source registries including PulseMCP, Smithery, Glama, and LobeHub.