Replace hardcoded list of popular models with dynamic catalog from getCatalogModels().
Enable searching by model name or description and include model descriptions in results. Improves discoverability and maintains up-to-date model list without manual updates. Database files updated accordingly.
Add `ensureOllamaAvailable` function to lib/ollama.ts for checking and waiting for Ollama server health with configurable timeout.
Integrate the check in GET /api/ollama/models route with 3s timeout to prevent UI blocking, and in POST /api/ollama/pull route with 20s timeout to ensure availability before pulling models.
Update ChatContainer to always render ModelDropdown for better UX when no model is selected.
Removed dependencies on MCPCacheRepository and PulseMCPService from MCP routes to simplify installation and listing logic, avoiding reliance on potentially stale cached data and streamlining the codebase. Updated available-for-chat filter to include MCPs without explicit status.
Add support for Model Context Protocol (MCP) tools in the Ollama chat API. Detect model capabilities to use native tools or prompt engineering. Handle tool calls, execute them via MCPExecutor, and inject results back into the conversation for seamless tool-assisted responses. This enhances chat functionality by enabling external tool integration.
- Support installing MCP servers from GitHub repos by converting 'github:user/repo' to 'git+https://github.com/user/repo.git' to avoid SSH issues and improve npm compatibility
- Display available tools in MCPCard with badges, showing up to 6 tools and total count for better visibility of server capabilities
- Remove unused onViewDetails prop and make uninstall button full-width for improved UI consistency
- Add /mcp-servers/ to .gitignore for MCP server installations
- Extend mcp_installations table with status, logs, and validation fields
- Refactor MCP install route to support custom configs, enable logs, and async installation
- Improve MCPSyncService with promise-based syncing to prevent concurrent operations
This update improves MCP management by adding robust installation tracking, custom configuration support, and better error handling for a more reliable user experience.
- Updated imports to use PulseMCPSearchParams and PulseMCPService instead of DeepNLP equivalents
- Replaced DeepNLP API calls with PulseMCP API endpoints and parameters
- Modified cached MCP transformation to manual mapping for PulseMCP data format
- Enhanced error handling and retry logic in server tools retrieval
- Added PulseMCP API base URL constant for centralized configuration
This change switches the underlying MCP provider from DeepNLP to PulseMCP, ensuring compatibility with the new API while maintaining existing functionality for MCP listing and tool retrieval.
Add comprehensive CSS styles for the MCP Marketplace interface, including card hover effects, responsive grid layout, modal backdrop blur, category badges, rating stars, install buttons, and loading animations. Also include custom scrollbar styling for the model selector to enhance user experience and visual consistency.
- Add currentConversationId prop to ChatContainer for loading selected conversations
- Implement logic to load messages when conversation is selected and clear when deselected
- Improve user message rendering with proper whitespace handling
- Enhance MarkdownRenderer with content normalization and adjusted spacing for better readability
- Add paragraph styling to markdown output for improved layout
This enables users to switch between conversations seamlessly and provides cleaner message display.
- Update .gitignore to exclude database files, extensions, and .cursor directory
- Enhance chat streaming API with safe enqueue and close functions to prevent errors
- Integrate ConversationSidebar component into ClientHome for better UI
- Add dependencies like better-sqlite3 and esbuild packages for database functionality
This commit introduces database persistence for conversations, improves error handling in streaming responses, and adds a sidebar for managing chats, enhancing overall app robustness and user experience.
Evandro Fonseca Junior