NotebookLM + GitHub Copilot + MCP

GitHub Copilot + NotebookLM + MCP = grounded AI engineering inside VS Code.

Turn GitHub Copilot into a source-grounded AI engineering agent using NotebookLM + MCP. Stop hallucinating architecture. Start citing documents.

Unofficial community project. Not affiliated with Google, GitHub, Microsoft, or OpenAI. A Claude Code Skills alternative for GitHub Copilot users.

---

Other languages: Español

---

Why this exists

GitHub Copilot is excellent at writing code. It is not designed to reason over your private documents — customer proposals, architecture specs, meeting notes, vendor docs.

Claude Code Skills solves this for Anthropic users. This project solves the same problem for GitHub Copilot users, using NotebookLM as the document intelligence layer and MCP as the integration protocol.

The result: Copilot Chat can query your NotebookLM notebooks directly, ground answers in your sources, and cite them — all inside VS Code.

---

Architecture

flowchart TD
    A[Developer in VS Code] -->|types question| B[GitHub Copilot Chat\nAgent Mode]
    B -->|MCP tool call over stdio| C[notebooklm-mcp\nvia npx]
    C -->|browser automation| D[Google NotebookLM]
    D -->|source-grounded answer\nwith citations| C
    C -->|structured response| B
    B -->|cited answer| A

    subgraph "Your Documents"
        E[Architecture Specs]
        F[Proposals & RFPs]
        G[Meeting Notes]
        H[Vendor Docs]
        E & F & G & H --> D
    end

Loading

Other supported clients (OpenCode, Cursor) connect to the same notebooklm-mcp server using the same stdio transport. See docs/mcp-clients.md for a full comparison.

---

What changed in v0.4

v0.4 evolves the toolkit from an enterprise reference kit into a demo-ready, easy-to-configure product that teams can try and adopt in minutes.

Area	What's new
Setup wizard	npm run setup:wizard — generates MCP config for VS Code, OpenCode, Cursor or generic clients
Config generator	Non-interactive npm run setup:vscode/opencode/cursor commands
Doctor command	npm run doctor — checks Node.js, npm, Chrome, client configs and package scripts
Demo kit	2-min, 5-min and 15-min scripts, sample outputs, recording checklist in demo/
Integration recipes	GitHub Issues and Azure DevOps prompt-driven workflows in integrations/
Sample sources	Fictional demo documents for NotebookLM in sample-sources/
Output schemas	JSON Schema 2020-12 definitions for ADRs, backlog, risk register and proposals
Output format prompts	Structured prompt packs in prompt-packs/output-formats/
Adoption maturity model	5-level adoption framework in docs/adoption-maturity-model.md
GitHub Pages site	Static docs site in site/ with _config.yml
Improved validation	check:schemas, check:site added to check:repo

See CHANGELOG.md for the full list.

---

What changed in v0.3

v0.3 adds enterprise rollout guidance, security hardening, governance templates, multi-notebook workflows, team prompt packs and an evaluation framework for source-grounded AI engineering.

Area	What's new
Security hardening	Threat model, privacy guide, compliance considerations, browser profile security
Governance	Policy templates, approved sources policy, MCP server governance, team adoption playbook
Multi-notebook workflows	Query patterns across multiple notebooks, conflict resolution, source triangulation
Team prompt packs	Ready-to-use prompts for platform engineering, cloud architecture, presales, security, delivery and developer enablement
Evaluation framework	Scoring rubrics for architecture answers, presales outputs, code generation, and security reviews
Security checklists	Operational checklists for MCP server approval, source approval, browser profile management, and enterprise rollout
Quality validation	New scripts: check:links, check:recipes, check:prompts, check:frontmatter, docs:index

See CHANGELOG.md for the full list.

---

Try it in 5 minutes

Prerequisites: GitHub Copilot, VS Code, Node.js 18+, Chrome stable, Google account with NotebookLM access.

git clone https://github.com/davidop/notebooklm-github-copilot.git
cd notebooklm-github-copilot
npm install
npm run doctor          # check your environment
npm run setup:wizard    # generate VS Code / Cursor / OpenCode config

1. Open .vscode/mcp.json — click the Start CodeLens to launch the MCP server.
2. Open Copilot Chat → select Agent mode → enable notebooklm tools.
3. Authenticate once:

   Use the NotebookLM MCP server to run setup_auth. Open the browser visibly so I can log in.
   

4. Verify it works:

   Use NotebookLM to list my available notebooks and confirm whether I am authenticated.
   

See docs/setup.md for a complete setup walkthrough.

---

Guided setup

The setup wizard generates MCP client configuration files without manual JSON editing:

npm run setup:wizard                                          # interactive
npm run setup:vscode                                          # VS Code / GitHub Copilot
npm run setup:opencode                                        # OpenCode
npm run setup:cursor                                          # Cursor
npm run setup:wizard -- --client vscode --version 0.1.0      # pinned version
npm run setup:wizard -- --client cursor --force               # overwrite existing

See docs/setup-wizard.md for full usage.

---

Doctor command

Check your local environment before starting:

npm run doctor

Checks Node.js version, npm/npx, repository structure, client config files, package scripts and optional Chrome detection. Provides remediation suggestions for each issue.

See docs/doctor.md for details.

---

How it works

1. MCP bridge — .vscode/mcp.json wires VS Code to notebooklm-mcp via the Model Context Protocol stdio transport.
2. Copilot instructions — .github/copilot-instructions.md and .github/instructions/ teach Copilot when and how to call NotebookLM tools.
3. NotebookLM as RAG — notebooklm-mcp drives a local Chrome session to query your notebooks, returning source-grounded answers with citations.
4. Recipes and prompts — Pre-built prompt patterns for ADRs, architecture reviews, presales, and more.

No documents leave your Google account. No NotebookLM code is vendored in this repo. The MCP server is managed by the notebooklm-mcp community package.

---

Copilot + NotebookLM MCP vs Claude Code Skills

Feature	This project	Claude Code Skills
AI assistant	GitHub Copilot	Claude Code
Editor	VS Code (any plan)	Terminal / any editor
Document grounding	NotebookLM via MCP	Anthropic RAG tools
Source citations	✅ via NotebookLM	✅
Notebook management	Google NotebookLM UI	Custom
Protocol	MCP (stdio/HTTP)	Custom skill API
Copilot Business/Enterprise	✅ (with org MCP policy)	❌
Setup complexity	Low (npx + Chrome)	Medium
Offline use	❌ (requires Google)	Depends on setup
Cost	Copilot + NotebookLM plans	Claude subscription

---

Use cases

Use case	What Copilot does
Architecture Decision Records	Queries prior decisions, generates ADR drafts grounded in your docs
Presales proposals	Reuses existing proposal language, cites past wins
RFP analysis	Extracts requirements, maps to capabilities from your knowledge base
Azure architecture	Generates Bicep/Terraform from vendor docs loaded into NotebookLM
Meeting notes → backlog	Turns action items from uploaded meeting notes into user stories
Code from vendor docs	Generates code aligned to vendor specifications in your notebooks
Technical documentation	Produces consistent docs by reusing your existing templates and style guides

---

Recipes

Step-by-step workflows for common engineering tasks:

Recipe	Description
Generate an ADR	Create an Architecture Decision Record from prior decisions in NotebookLM
Azure architecture	Generate Azure architecture from vendor or customer docs
Compare proposals	Compare two proposals against requirements
Backlog from meeting notes	Turn uploaded meeting notes into a sprint backlog
Bicep from docs	Generate Bicep templates grounded in architecture specs
Terraform from docs	Generate Terraform modules grounded in architecture specs
Review an RFP	Analyse an RFP against your capability library

---

GitHub and Azure DevOps integration recipes

Prompt-driven workflows for GitHub Issues and Azure DevOps:

GitHub Issues:

Recipe	Description
Create issues from NotebookLM	Turn NotebookLM outputs into GitHub Issues
Triage issues with NotebookLM	Ground issue triage in your knowledge base
Generate PR description	Generate grounded pull request descriptions
Generate release notes	Create release notes from NotebookLM sources
ADR to GitHub Issues	Convert ADRs into trackable GitHub Issues

Azure DevOps:

Recipe	Description
Work items from meeting notes	Create work items from uploaded meeting notes
Epics, features, user stories	Generate backlog hierarchy from requirements
Acceptance criteria	Generate grounded acceptance criteria
Sprint planning	Plan sprints using NotebookLM sources
Delivery risk review	Review delivery risks with NotebookLM context

---

Demo kit

Everything you need to record a compelling demo without confidential data:

• Demo README — Overview and setup assumptions
• 2-minute script — Quick intro demo
• 5-minute script — Standard team demo
• 15-minute script — Deep-dive technical demo
• Sample prompts — Ready-to-paste prompts
• Recording checklist — Pre/during/post recording guide

---

Sample sources

Fictional demo documents safe to upload to NotebookLM for demos:

Disclaimer: These files are fictional demo sources and do not represent a real customer.

File	Description
cloud-modernization-brief.md	Fictional cloud modernization brief
architecture-principles.md	Fictional architecture principles
security-requirements.md	Fictional security requirements
customer-meeting-notes.md	Fictional meeting notes
vendor-implementation-guide.md	Fictional vendor guide
previous-proposal-summary.md	Fictional proposal summary

---

Output schemas

JSON Schema definitions for structured AI outputs:

Schema	Description
adr.schema.json	Architecture Decision Record
architecture-review.schema.json	Architecture review report
backlog-items.schema.json	Sprint backlog items
presales-proposal.schema.json	Presales proposal
risk-register.schema.json	Risk register

---

Adoption maturity model

A five-level framework for adopting NotebookLM + MCP across teams:

Level	Stage
0	Ad hoc local experiments
1	Individual developer setup
2	Team-shared prompts and recipes
3	Governed enterprise rollout
4	Evaluated and continuously improved workflows

See docs/adoption-maturity-model.md for the full model.

---

GitHub Pages documentation

The site/ folder contains a static documentation site compatible with GitHub Pages:

• Getting started
• Client configuration
• Recipes overview
• Security
• Governance
• FAQ

---

Security and privacy

• No secrets in NotebookLM. Never upload API keys, credentials, connection strings, or regulated personal data to Google NotebookLM.
• Authentication is local. The MCP server stores a Chrome profile on your machine. It is not committed to this repository.
• Google's data processing applies. Documents uploaded to NotebookLM are subject to Google's terms of service and privacy policy.
• Scope documents carefully. Only upload documents your organization has approved for cloud storage and AI processing.

See security/threat-model.md and SECURITY.md for full details.

---

Enterprise rollout

For organizations deploying this to multiple developers:

1. Enable MCP in Copilot policy — Copilot Business/Enterprise requires explicit org policy for MCP servers.
2. Pin the MCP server version — Set a specific notebooklm-mcp@x.y.z in .vscode/mcp.json rather than @latest.
3. Document approved notebooks — Maintain an internal registry of approved NotebookLM notebooks per team.
4. Train on data classification — Ensure engineers know what is safe to upload.

See docs/enterprise-rollout.md for a full checklist.

---

Security hardening

See dedicated documentation for security, privacy and compliance:

• Security hardening guide
• Privacy and data handling
• Compliance considerations
• Browser profile security
• MCP threat model

---

Governance templates

Generic, reusable policy templates for enterprise teams adopting AI-assisted engineering:

• AI-assisted engineering policy
• Approved sources policy
• NotebookLM usage policy
• MCP server governance
• Team adoption playbook

See governance/ for the full collection.

---

Evaluation framework

Scoring rubrics and scenarios for reviewing the quality of AI-assisted engineering outputs:

• Source grounding scorecard
• Architecture answer evaluation
• Presales output evaluation
• Code generation evaluation

See evals/ for all rubrics and scenarios.

---

Multi-notebook workflows

Work with multiple specialized notebooks in a single workflow — combining customer requirements, internal standards, vendor docs, and prior proposals:

• Multi-notebook workflows guide
• Multi-notebook research recipe
• Source triangulation
• Conflict resolution between sources

---

Team prompt packs

Copy-paste-ready prompt collections for specific teams and roles:

Team	Prompt pack
Platform engineering	prompt-packs/team/platform-engineering.md
Cloud architecture	prompt-packs/team/cloud-architecture.md
Presales	prompt-packs/team/presales.md
Security	prompt-packs/team/security.md
Delivery management	prompt-packs/team/delivery-management.md
Developer enablement	prompt-packs/team/developer-enablement.md

---

Limitations

• Browser automation fragility — notebooklm-mcp uses browser automation against the NotebookLM UI. Google UI changes can break it until the package is updated.
• No offline mode — Requires an active Google session and internet connection.
• Authentication is per-developer — No centralized service account for NotebookLM (Google does not expose a public API).
• NotebookLM source limits — Notebooks have source count and size limits imposed by Google.
• Codespaces caveat — Interactive browser authentication works best from a local VS Code environment, not GitHub Codespaces.
• Auditability gaps — NotebookLM does not expose audit logs. Track notebook queries through your coding agent's conversation history.
• No centralized policy enforcement — Governance templates are advisory. Enforcement depends on organizational controls.
• Evaluation framework is manual — The evals/ rubrics require human reviewers. There is no automated grading.

---

Disclaimer

This is an unofficial community project. It is not affiliated with, endorsed by, or supported by Google, GitHub, Microsoft, or OpenAI. Use it at your own risk.

The notebooklm-mcp server is a third-party package. Review its license and security posture before enterprise deployment.

---

Repository layout

.github/
  copilot-instructions.md
  instructions/
  workflows/
  ISSUE_TEMPLATE/
  PULL_REQUEST_TEMPLATE.md
.vscode/
  mcp.json
  settings.json
  extensions.json
.devcontainer/
  devcontainer.json
  README.md
assets/
  social-preview.svg
clients/
  vscode/
    mcp.json
    copilot-instructions.example.md
    README.md
  opencode/
    opencode.jsonc
    agent-researcher.md
    agent-architect.md
    agent-presales.md
    README.md
  cursor/
    mcp.json
    rules/
    README.md
docs/
  setup.md
  usage.md
  troubleshooting.md
  operating-model.md
  enterprise-rollout.md
  faq.md
  mcp-clients.md
  devcontainer.md
examples/
  prompts.md
  adr-from-notebooklm/
  architecture-review/
  presales-proposal/
  code-from-vendor-docs/
  meeting-notes-to-backlog/
prompts/
recipes/
security/
  threat-model.md
governance/
  README.md
  ai-assisted-engineering-policy.md
  approved-sources-policy.md
  notebooklm-usage-policy.md
  mcp-server-governance.md
  prompt-and-output-review-policy.md
  team-adoption-playbook.md
  exception-request-template.md
checklists/
  README.md
  security/
    mcp-server-approval-checklist.md
    notebooklm-source-approval-checklist.md
    customer-data-review-checklist.md
    browser-profile-checklist.md
    prompt-injection-review-checklist.md
    enterprise-rollout-security-checklist.md
evals/
  README.md
  source-grounding-scorecard.md
  architecture-answer-evaluation.md
  presales-output-evaluation.md
  code-generation-from-docs-evaluation.md
  security-review-evaluation.md
  rubric-template.md
  scenarios/
    adr-generation.md
    azure-architecture-review.md
    rfp-review.md
    security-threat-model.md
    code-from-vendor-docs.md
prompt-packs/
  README.md
  team/
    README.md
    platform-engineering.md
    cloud-architecture.md
    presales.md
    security.md
    delivery-management.md
    developer-enablement.md
  output-formats/
    README.md
    json-adr.md
    json-backlog.md
    json-risk-register.md
    markdown-executive-summary.md
    markdown-architecture-review.md
marketing/
  v0.3-launch.md
  v0.4-launch.md
integrations/
  github/
    README.md
    recipes/
  azure-devops/
    README.md
    recipes/
demo/
  README.md
  demo-script-2-min.md
  demo-script-5-min.md
  demo-script-15-min.md
  sample-prompts.md
  sample-notebook-sources.md
  recording-checklist.md
  sample-outputs/
site/
  index.md
  getting-started.md
  clients.md
  recipes.md
  security.md
  governance.md
  evaluations.md
  demo.md
  faq.md
sample-sources/
  README.md
  cloud-modernization-brief.md
  architecture-principles.md
  security-requirements.md
  customer-meeting-notes.md
  vendor-implementation-guide.md
  previous-proposal-summary.md
schemas/
  README.md
  adr.schema.json
  architecture-review.schema.json
  backlog-items.schema.json
  presales-proposal.schema.json
  risk-register.schema.json
  examples/
templates/
scripts/
  validate.mjs
  smoke-test.mjs
  setup-wizard.mjs
  generate-client-config.mjs
  doctor.mjs
  check-schemas.mjs
  check-site.mjs
  release-notes.mjs
_config.yml
CHANGELOG.md
CONTRIBUTING.md
ROADMAP.md
SECURITY.md
SUPPORT.md
README.es.md
package.json

---

Recommended adoption path

New users:

1. Run npm run doctor to check your environment
2. Run npm run setup:wizard to generate your MCP config
3. Pick a recipe from recipes/ that matches your workflow
4. Review docs/security-hardening.md before uploading documents

Teams adopting for the first time:

1. Read the adoption maturity model
2. Read the team adoption playbook
3. Complete the enterprise rollout security checklist
4. Choose prompt packs from prompt-packs/team/
5. Use the demo kit to onboard the team

Enterprise / governed deployments:

1. Adapt the governance templates to your organization
2. Complete the MCP server approval checklist
3. Set up approved notebook registry per approved-sources-policy.md
4. Train developers on privacy and data handling
5. Establish evaluation cadence using evals/

---

Contributing

See CONTRIBUTING.md. All contributions welcome — especially new recipes and real-world examples.

Changelog

See CHANGELOG.md.

Roadmap

See ROADMAP.md.

License

MIT. See LICENSE.