- Updated extension.ts MCP server definition - Updated server.ts config name - Added auto-growing textarea to web interface - Tool now referenced as HumanAgentMCP/HumanAgent_Chat (no space) - Fixes tool reference issues in .agent.md files
HumanAgent MCP
Forces GitHub Copilot to chat with you before acting. Stops runaway agents, reduces wasted API calls, lets you manage multiple workspaces from one interface.
Installation
- Install from VS Code Marketplace
- Copilot automatically gets the
HumanAgent_Chattool - Done - no configuration needed
How to Use
Basic Workflow
- Ask Copilot to do something - Copilot will use the HumanAgent_Chat tool
- Chat panel opens - Green dot = connected, shows Copilot's message
- You respond - Type your answer, click Send (or use Quick Replies)
- Copilot continues - Gets your response and proceeds with the task
VS Code Interface
Chat Panel (left sidebar):
- Green dot = connected to server
- Red dot = disconnected (auto-reconnects)
- Quick Replies = common responses like "Yes Please Proceed"
- Text input = always enabled, send button only active when Copilot is waiting
Cog Menu (⚙️):
- Show Status - check server state
- Name This Chat - set session name
- Open Web View - manage all workspaces in browser
- Configure MCP - start/stop/restart server
- Report Issue / Request Feature - GitHub links
Web Interface
Open from cog menu → Open Web View
Access all workspace chats in one browser tab at http://localhost:3737/HumanAgent
- See all conversations
- Switch between workspaces
- Append reminders to your responses
Connection Status
- Auto-reconnects with exponential backoff (1s → 2s → 4s → 8s → 16s → 30s max)
- Manually starting server resets reconnection immediately
- No timeout - will retry forever
Troubleshooting
Red dot / disconnected:
- Cog menu → Configure MCP → Start Server
- Check VS Code Output panel for errors
Server won't start:
- Check port 3737 not in use:
lsof -i :3737 - Restart VS Code
Copilot not using the tool:
- Tool registers automatically on startup
- Try: "Use HumanAgent_Chat to discuss this with me"
Development
Press F5 to debug - dev mode uses port 3738, production uses 3737. No conflicts.
Privacy & Telemetry
This extension collects anonymous usage data to help improve the product:
What we track:
- Extension activation/deactivation
- Feature usage (chat opened, messages sent/received)
- Error diagnostics (error types, not content)
- Session metrics (message counts, not content)
- Extension version, VS Code version, OS platform
- Days since installation
What we DON'T track:
- ❌ Your message content
- ❌ Your name, email, or any personal data
- ❌ Workspace paths or file names
- ❌ Any identifiable information
Your privacy:
- Respects VS Code's telemetry setting
- To disable: Settings → Telemetry → Level → Off
- Fully GDPR compliant
- Uses Google Analytics 4 for anonymous metrics
Why telemetry?
- Helps us understand which features are used
- Identifies bugs and errors to fix
- Measures engagement and retention
- Guides future development priorities
For questions: GitHub Issues
More Info
See README-Additional.md for technical details
Demo
Complete demonstration of the HumanAgent MCP extension in action - showing real-time human-AI collaboration
