MCP Marketplace and Search Provider Expansion 🔍

From June through July, LobeHub focused on two things: making tools easier to discover and making search more flexible. The new MCP Marketplace brings one-click plugin installation to desktop, while expanded search providers and authentication options give teams more choice in how they work.

MCP Marketplace: discover and install tools faster

Finding and installing MCP plugins used to require manual setup. Now the MCP Marketplace brings one-click installation directly to the desktop app. Browse available tools, see what they do, and add them to your workspace without leaving the interface.

This makes the plugin ecosystem more accessible—especially for teams who want to extend their Agents without diving into configuration files or command-line steps.

More ways to search and authenticate

Search coverage now includes Brave, Google PSE, and Kagi as built-in options, plus Vertex AI Google Search Grounding for grounded generation workflows. You can choose the provider that fits your accuracy, speed, and privacy needs.

On the authentication side, Amazon Cognito and Google SSO are now available as identity providers. Teams already using these systems can connect LobeHub without adding another set of credentials. Page access protection is also included, so you can control who sees what based on their sign-in status.

Runtime improvements

This cycle also brings v0 (Vercel) provider support and an analytics event-tracking framework for better visibility into usage patterns.

Improvements and fixes

• Refined mobile model selector layout and text overflow handling
• Improved loading animation transitions
• Optimized Gemini thinkingBudget configuration and reasoning_effort parameter handling
• Added Browserless blockAds and stealth parameters for cleaner search results
• Fixed Mermaid rendering issues in Firefox
• Improved multi-monitor window opening behavior on desktop
• Fixed theme-related edge cases and optimized chunked loading
• Added transition animation toggle for response rendering