Skip to content

mcp-clients.md

Latest commit

 

History

History
149 lines (110 loc) · 5.51 KB

mcp-clients.md

File metadata and controls

149 lines (110 loc) · 5.51 KB
Raw

MCP clients comparison

This document compares the supported MCP client integrations for the NotebookLM + MCP toolkit.

Architecture

flowchart LR
    A[GitHub Copilot / VS Code] --> M[NotebookLM MCP Server]
    B[OpenCode] --> M
    C[Cursor] --> M
    D[Other MCP Clients] --> M
    M --> N[Google NotebookLM]
    N --> M
    M --> A
    M --> B
    M --> C
    M --> D
Loading

All supported clients connect to the same notebooklm-mcp server process via the stdio MCP transport. The server handles browser automation to query Google NotebookLM and returns grounded, cited answers.

Client comparison table

GitHub Copilot (VS Code) OpenCode Cursor Generic MCP client
Config file .vscode/mcp.json ~/.config/opencode/config.jsonc .cursor/mcp.json or ~/.cursor/mcp.json Varies by client
MCP transport stdio stdio stdio stdio or HTTP
Agent / chat mode Copilot Chat (Agent mode) OpenCode agent session Cursor AI chat Depends on client
Recommended use cases VS Code-native workflows, code generation, ADRs Terminal-first workflows, scripting, automation In-editor chat, code generation from docs CI pipelines, custom tooling
Strengths Deep VS Code integration; repository instructions via .github/; Copilot Business/Enterprise policy support Lightweight; fast terminal workflow; scriptable agent sessions Cursor Rules for fine-grained control; strong code generation Flexible; can integrate into any MCP-compatible tool
Limitations Requires VS Code; Copilot plan required Less mature MCP ecosystem than VS Code; terminal-only Cursor-specific rules format; requires Cursor subscription No standardized instruction/rules mechanism; integration varies
Auth requirement GitHub Copilot + Google account Google account (local Chrome) Google account (local Chrome) Google account (local Chrome)
Offline support ❌ ❌ ❌ ❌
Container support Editing only (no live MCP in headless) Editing only Editing only Depends

GitHub Copilot in VS Code

Config: .vscode/mcp.json

Strengths:

  • First-class integration with VS Code and GitHub Copilot.
  • Repository instructions (.github/copilot-instructions.md) are automatically loaded by Copilot.
  • Workspace-level config is easy to commit and share across a team.
  • Compatible with Copilot Business and Enterprise (with org MCP policy enabled).
  • CodeLens start/stop controls for the MCP server directly in the config file.

Limitations:

  • Requires VS Code and a GitHub Copilot subscription.
  • Browser authentication requires a local VS Code environment (not Codespaces).

Setup: See clients/vscode/README.md

OpenCode

Config: ~/.config/opencode/config.jsonc

Strengths:

  • Terminal-first workflow, no editor required.
  • Well-suited for scripting, automation, and CI-adjacent workflows.
  • Simple JSONC configuration format.
  • Supports custom agent personas via instruction files.

Limitations:

  • Less mature MCP ecosystem compared to VS Code.
  • No persistent repository-level instructions mechanism equivalent to .github/copilot-instructions.md.
  • Requires a local terminal environment for browser authentication.

Setup: See clients/opencode/README.md

Cursor

Config: .cursor/mcp.json (workspace) or ~/.cursor/mcp.json (user)

Strengths:

  • Strong in-editor AI chat and code generation.
  • Cursor Rules (.mdc files) provide fine-grained, file-type-scoped instructions.
  • Workspace-level config can be committed alongside project files.
  • Flexible rule targeting via glob patterns.

Limitations:

  • Requires a Cursor subscription.
  • Rules format (.mdc) is Cursor-specific and not portable to other clients.
  • Browser authentication requires a local Cursor environment.

Setup: See clients/cursor/README.md

Generic MCP-compatible clients

Any MCP-compatible client can connect to notebooklm-mcp using the stdio transport.

Configuration pattern:

{
  "mcpServers": {
    "notebooklm": {
      "command": "npx",
      "args": ["-y", "notebooklm-mcp@latest"],
      "env": {
        "NOTEBOOKLM_ACCOUNT": "work"
      }
    }
  }
}

Strengths:

  • Any MCP-compatible tool can use this configuration.
  • Useful for custom tooling, CI pipelines, or experimental agent frameworks.

Limitations:

  • No standardized instruction or rules mechanism across clients.
  • Integration quality varies significantly by client.
  • Browser authentication still requires a local machine with Chrome.

Choosing the right client

Scenario Recommended client
Day-to-day VS Code development GitHub Copilot in VS Code
Terminal-first or scripting workflows OpenCode
Cursor users wanting grounded code generation Cursor
Custom tooling or CI integration Generic MCP client
Presales or architecture documentation Any — use the agent instruction files

Common limitations across all clients

  • Browser automation — notebooklm-mcp uses local Chrome. Google UI changes can break it until the package is updated.
  • No offline mode — NotebookLM is a cloud service.
  • Per-developer authentication — No shared service account available.
  • NotebookLM source limits — Google imposes notebook source count and size limits.
  • Not official integrations — All configurations rely on the community notebooklm-mcp package.