- Notifications
You must be signed in to change notification settings - Fork2.7k
Description
Describe the feature or problem you'd like to solve
The remote GitHub MCP server athttps://api.githubcopilot.com/mcp/
currently exposes many dozen tools by default with no documented way to limit or configure which toolsets are available. While the Docker-based server supports toolset configuration via theGITHUB_TOOLSETS
environment variable, the remote server lacks equivalent functionality.

This creates several problems:
- AI agents can become overwhelmed by too many tool options, leading to poor tool selection
- Some MCP hosts (like Cursor) only send the first 40 tools to agents, meaning 60+ tools are inaccessible
- Users who only need specific GitHub functionality (e.g., read-only code review, issue management only) must expose unnecessary and potentially risky tools
- Organizations wanting to limit scope for security/compliance have no granular control beyond the all-or-nothing
X-MCP-Readonly
header
Proposed solution
Add support for toolset configuration in the remote server via headers or URL parameters, similar to how the Docker version uses environment variables. This could be implemented as:
Option 1: Header-based configuration
{"headers": {"Authorization":"Bearer TOKEN","X-MCP-Toolsets":"context,repos,issues,pull_requests" }}
Option 2: URL parameter
{"url":"https://api.githubcopilot.com/mcp/?toolsets=context,repos,issues"}
Option 3: Dynamic toolset discovery (already supported in Docker)
{"headers": {"X-MCP-Dynamic-Toolsets":"true" }}
This would benefit users by:
- Improving AI agent accuracy through reduced tool confusion
- Ensuring all desired tools fit within host limitations (40-tool limit)
- Providing security-conscious organizations with granular access control
- Maintaining feature parity between Docker and remote server implementations
Example prompts or workflows (for tools/toolsets only)
Code Review Workflow - Enable only:
context,repos,pull_requests
- "Review the latest PR and check if it follows our coding standards"
- "Show me all open PRs that modify authentication logic"
- "What changes were made in PRIssue: Fix incorrect VS Code installation URL (both buttons point to Insiders) #123?"
Issue Triage Workflow - Enable only:
context,issues
- "List all high-priority bugs from this week"
- "Create an issue for the bug we just discussed"
- "Show me issues labeled 'good-first-issue' that are unassigned"
Security Audit Workflow - Enable only:
context,code_security
- "Show me all critical security alerts in our repositories"
- "What dependencies have known vulnerabilities?"
- "Check if any secrets were exposed in recent commits"
Documentation Workflow - Enable only:
context,repos
(read-only mode)- "Find all README files across our organization's repos"
- "Show me the API documentation in the docs folder"
- "What examples are provided for authentication?"
CI/CD Monitoring - Enable only:
context,actions
- "Which workflows failed in the last 24 hours?"
- "Show me the deployment pipeline configuration"
- "Re-run the failed test suite from this morning"
Additional context
- The Docker implementation already demonstrates this is technically feasible and valuable
- The documentation athttps://github.blog/ai-and-ml/generative-ai/a-practical-guide-on-how-to-use-the-github-mcp-server/ mentions toolset configuration but doesn't provide complete implementation details for the remote server
- With the rapid adoption of MCP across different AI coding assistants (Cursor, Claude, VS Code), having consistent configuration options across deployment methods would improve developer experience
- This feature would align with the principle of least privilege, allowing organizations to grant only the minimum necessary access for specific workflows