A curated collection of Model Context Protocol (MCP) toolset profiles for GitHub Copilot Chat (Agent mode), designed to make AI assistance explicit, safe, and context-appropriate.
This repository contains:
default.toolsets.jsonc— Five ready-to-use MCP toolset profiles that control which tools the Copilot Agent can accesstoolset-profiles.md— Complete guide on the profiles themselves: their philosophy, security model, how to use them, and enterprise deployment considerationsmcp-troubleshooting.md— Comprehensive troubleshooting guide with step-by-step diagnostics, common pitfalls, workarounds, and Windows-specific tipsmcp-troubleshooting-quick.md— Fast-reference checklist for common MCP issues
When you enable GitHub Copilot in Agent mode, you're giving an AI assistant access to a powerful toolkit. The challenge? Not every task needs—or should have—every capability.
MCP toolset profiles solve this by letting you:
- Match capability to context: Use a minimal, safe set for everyday edits; switch to a comprehensive development toolkit when building and testing; isolate destructive operations in a dedicated profile.
- Make permissions explicit: The available tools are declared in code, reviewed like code, and shared like code—no surprises.
- Improve safety by default: Keep risky operations (file deletions, memory graph modifications) in a separate, opt-in profile that you switch to intentionally and then switch away from.
- Enhance team collaboration: Share consistent, named profiles across your team so everyone knows what the Agent is allowed to do.
Think of it as role-based access control for your AI pair programmer.
Minimal, safe defaults for everyday chat and edits
Perfect for quick questions, documentation updates, and small code changes. Includes only non-destructive tools: search, edit, new, usages, problems, and a few others. No filesystem writes, no deletions.
Comprehensive, non-destructive development toolkit
Your daily driver for serious development work. Includes code editing, testing, notebooks, Python environments, runtime inspection (Console Ninja), filesystem operations, documentation fetching (Context7), and non-destructive memory graph operations. Still excludes memory deletions to keep things safe by default.
Notebooks, data exploration, and safe reads
Focused on data science workflows: execute notebooks, manage Python packages, inspect environments with Pylance, and read files safely. Excludes filesystem writes and destructive operations—perfect for exploration without side effects.
Isolated destructive operations (USE WITH CAUTION)
The "break glass in case of emergency" profile. Contains tools for deleting memory graph entities and writing/moving files. Use this profile only for the specific task that requires destructive actions, then switch back immediately.
Hard constraints: propose-only mode
Extremely limited toolset for compliance reviews or when you want the Agent to suggest changes without executing anything. Only search, changes, edit, and new.
-
Copy the profile file
Placedefault.toolsets.jsoncin your VS Code user prompts directory:- Windows:
%APPDATA%/Code/User/prompts/default-tools.toolsets.jsonc - macOS:
~/Library/Application Support/Code/User/prompts/default-tools.toolsets.jsonc - Linux:
~/.config/Code/User/prompts/default-tools.toolsets.jsonc
- Windows:
-
Reload VS Code
RunDeveloper: Reload Windowfrom the Command Palette. -
Switch to Agent mode
Open GitHub Copilot Chat and select Agent mode (not Ask mode). -
Select a profile
Choose your toolset by name in the Agent interface (e.g.,graph-basicorgraph-development). -
Verify
Ask: "What tools are enabled right now?" to confirm the profile is active.
| Task | Profile |
|---|---|
| Writing docs, small code edits | graph-basic |
| Day-to-day development, building, testing | graph-development |
| Data science, notebooks, exploring datasets | graph-data-analysis |
| Deleting files, removing memory graph entities | graph-demolition |
| Compliance reviews, propose-only mode | graph-none |
Golden rule: Start with graph-basic. Switch up only when you need the extra capability. Switch back when done.
This document explains:
- Why profiles exist: Safety, clarity, reproducibility, governance, and performance
- How they're structured: JSONC format, naming conventions, tool grouping
- Security model: Isolating destructive operations, keeping profiles lean, treating changes like code
- Enterprise deployment: Shared profiles, change control, permissions (with Windows NTFS examples), and the risks of uncontrolled profile modifications
- Prerequisites: Which tools require additional extensions or runtimes (Console Ninja, Docker, Python/Pylance, Context7)
Key insight: Profiles are code. Review them, version them, and control who can modify them—especially in regulated or enterprise environments.
When MCP servers don't start or tools don't appear, this guide walks you through:
- Configuration validation: Checking
mcpServerssettings, URLs, ports, command paths - Agent mode verification: Ensuring Copilot Chat is in the right mode
- Independent server testing: HTTP health checks, stdio command validation
- Log analysis: Where to find errors (Output, DevTools, MCP audit logs)
- Common issues: PATH problems, port conflicts, proxy interception, firewall blocks, TLS certificate issues, EPERM errors on Windows
- Platform-specific tips: Windows PATH configuration, development drives, WinHTTP proxy bypass, firewall rules
Includes a complete checklist and reset sequence for systematic debugging.
The TL;DR version: fast checks, common issues, Windows-specific commands (proxy, firewall), and when to escalate.
Perfect for quick troubleshooting or as a team reference card.
Critical warning: MCP profiles enable function calling from external sources. A single misconfiguration can grant the Agent unsafe capabilities.
- Treat profiles like code: Use pull requests, require reviews, maintain change logs
- Restrict write access: In shared/enterprise scenarios, make the canonical profile read-only for regular users
- Isolate destructive tools: Keep file deletions and memory graph modifications in a separate, clearly named profile
- Validate before deploying: Test profile changes in a non-production environment first
- Audit tool usage: Monitor MCP tool-call logs for unexpected capability use
See toolset-profiles.md for detailed governance recommendations, including Windows NTFS permission examples and change control workflows.
Some tools require additional setup:
- Console Ninja (runtime logs): Install the Console Ninja extension
- Docker (container commands): Install Docker Desktop and ensure Docker CLI is in PATH
- Python/Pylance (Python environments, code execution): Install the Python extension and Pylance, then select an interpreter
- Context7 (documentation fetching): Requires internet access
If tools don't appear or MCP servers fail to start:
- Verify you're in Agent mode (not Ask mode)
- Reload VS Code after copying/editing profiles
- Check that required extensions are installed
- Review logs:
View → Output → GitHub Copilot/GitHub Copilot Agent/MCP - For detailed diagnostics, see
mcp-troubleshooting.md
Windows users: Pay special attention to PATH configuration, proxy bypass settings, and firewall rules—all covered in the troubleshooting guide.
For teams and organizations:
- Version control: Keep this repository in your internal Git; distribute via onboarding docs
- Change control: Require pull requests and approvals for profile modifications
- Shared profiles (quick-and-dirty): Place a read-only canonical JSONC on a network share or internal repo; users copy it into their user-level settings
- Safer alternative: Use enterprise config management (Intune, Jamf, SCCM, Ansible) with code review and approvals
- CODEOWNERS: Gate risky profiles (demolition/admin) behind specific approvers
See the Enterprise Use section in toolset-profiles.md for detailed deployment patterns and NTFS permission examples.
This repository demonstrates:
- Context-aware AI assistance: Different tasks need different capabilities; profiles make the switch explicit and safe
- Safety by default: Destructive operations are isolated; you opt in intentionally and temporarily
- Team collaboration: Consistent, named profiles give everyone a shared understanding of what the Agent can do
- Enterprise readiness: Profiles can be versioned, reviewed, audited, and deployed with change control
- Practical troubleshooting: Real-world deployment challenges (PATH, proxies, firewalls, TLS) are documented with solutions
Whether you're a solo developer looking to organize your AI toolkit, a team lead standardizing Copilot usage, or an enterprise architect planning a controlled rollout, these profiles and guides provide a solid foundation.
If you're facing issues or want assistance deploying MCP servers and profiles in your environment, please get in touch:
📧 hello@graphtechnologies.xyz
🌐 https://graphtechnologies.xyz
We offer custom solutions, enterprise support, and advanced MCP integrations.
Author: Ryan Hughes
Company: Graph Technologies
Date: November 5th, 2025
The information in this repository is provided for educational and informational purposes only. All snippets, commands, and configuration examples are illustrative; they may be incomplete, out of date, or require adaptation for your environment.
The author and Graph Technologies provide this material "as is," without warranties of any kind, express or implied, and accept no liability for any damages or issues arising from the use or misuse of the contents.
- Always obtain appropriate reviews and approvals before enabling tools that can modify files, execute commands, or perform destructive actions.
- In regulated or enterprise environments, treat profile changes like code (PRs, change logs) and validate on non-production machines first.
- Nothing here constitutes security advice; consult your internal security/compliance teams for policy and approval requirements.
If you intend to deploy similar systems into production, ensure you have the appropriate expertise, perform a formal risk assessment, test changes in an isolated sandbox, and verify compliance with your organization's policies.