MCP Hub

Mantra v0.7.0 introduces the powerful MCP Hub, the central nervous system for your AI programming tools.

What is MCP Hub?

MCP Hub is a full implementation of the MCP (Model Context Protocol) open standard initiated by Anthropic. It aggregates multiple MCP services into a compliant Streamable HTTP server, acting as both a "traffic relay" and a "control panel":

• Unified Entry Point: Aggregates multiple MCP services (supporting all standard protocols like stdio, SSE).
• Transparent Takeover: MCP Hub can take over MCP configurations from tools like Claude Code and Cursor, allowing you to manage permissions for all AI tools in one place.
• Security Control: Unified management of service associations and tool enablement/disablement via a graphical interface.

Key Features

1. Service Management: Add, edit, and toggle MCP services (supporting all standard MCP protocols).
2. Environment Variable Management: Configure global or project-specific environment variables for MCP services, with sensitive info masking.
3. OAuth Credential Management: Securely store authentication tokens required for remote MCP services (e.g., Google Drive).
4. Status Monitoring: Real-time view of hub status, connected clients, and active services.

Configuration Takeover

The core workflow of MCP Hub is takeover: one-click import of MCP configurations from your existing AI tools, with no manual reconfiguration needed.

Supported Tools

Tool	User-level Config	Project-level Config
Claude Code	✅	✅ (local scope)
Cursor	✅	✅
Gemini CLI	✅	✅
Codex	✅	✅

Smart Merge Engine

During takeover, MCP Hub uses a three-tier classification strategy to handle configuration differences:

• New: Services that don't exist in the Hub, imported directly
• Updated: Existing services with configuration changes, prompts for update
• Conflict: Key fields are inconsistent, provides diff comparison and manual decision

Original configurations are automatically backed up before each takeover operation, supporting atomic recovery.

Project-Level Management

v0.8.2 added project-level MCP management capabilities:

• Sidebar Management Panel: Manage MCP service associations directly from the project context menu
• Scoped Config Import: Auto-scan and filter configurations based on project scope
• Enhanced Detection: Automatically identify new services, filter out existing ones

Tool Permission Management (Tool Policy)

Mantra implements industry-leading project-level granular tool permission control.

Core Capabilities

• Project-level Policies: Set different tool access permissions for different projects. For example, allow AI to modify files in "Project A" but only read them in "Project B".
• Precise Permission Management: Set permissions for individual tools (e.g., read_file, shell_execute):
  • Enable: Allow AI to use the tool.
  • Disable: Prohibit AI from using the tool; the Hub will intercept the call.
• Dynamic Takeover: By taking over configurations from other AI tools, Mantra can enforce these security policies even when you are using native Claude Code.
• Strict Mode: Filter services based on project context, ensuring only MCP services relevant to the current project are exposed.

MCP Roots Protocol

MCP Hub implements the MCP Roots protocol, supporting project-level configuration awareness:

• AI tools (Claude Code, Gemini CLI, Codex, Cursor) inform the current working directory via roots/list requests
• MCP Hub uses Longest Prefix Match (LPM) algorithm to automatically route to the corresponding MCP services based on project context
• No manual project switching needed — MCP Hub automatically detects which project you're working in

Built-in MCP Inspector

To help developers debug MCP services, Mantra includes a professional built-in Inspector:

• Real-time Logs: View JSON-RPC communication between AI models and MCP services.
• Manual Trigger: Manually invoke tools in the Inspector to verify that services return the correct context.
• Troubleshooting: Quickly identify connection timeouts, permission errors, or data format mismatches.

Backup & Restore

MCP Hub provides comprehensive backup and restore mechanisms to safeguard your configurations:

• Atomic Operations: Both takeover and restore operations are atomic, with automatic rollback on failure
• Version Retention: Retains the last 5 backup versions
• Integrity Verification: Automatic backup file integrity verification before restore
• Auto Cleanup: Expired backups are automatically cleaned up

System Tray Integration

To ensure the MCP Hub provides continuous service in the background, Mantra deeply integrates with the system tray:

• Background Execution: Mantra keeps running in the background when the main window is closed, maintaining MCP service availability.
• Quick Access: Open the Hub, Inspector, or quit the app directly from the tray menu.
• Status Notifications: Real-time notifications via tray icon changes for hub exceptions or interception events.

Streamable HTTP Compliance

MCP Hub complies with the MCP Streamable HTTP specification (2025-03-26 version):

• Unified Endpoint: /mcp supports POST, GET, and DELETE operations
• Session Management: Managed via the MCP-Session-Id header
• Origin Validation: Prevents cross-site request forgery
• Backward Compatibility: Also supports legacy /sse and /message endpoints

Next Steps

• Skills Hub - Learn about unified Skills management (symmetric design with MCP Hub)
• Project Management - Learn how to associate MCP services with projects
• Keyboard Shortcuts - Learn how to quickly toggle debug modes