* refactor: rewrite filesystem MCP server with new tool set
- Replace existing filesystem MCP with modular architecture
- Implement 6 new tools: glob, ls, grep, read, write, delete
- Add comprehensive TypeScript types and Zod schemas
- Maintain security with path validation and allowed directories
- Improve error handling and user feedback
- Add result limits for performance (100 files/matches max)
- Format output with clear, helpful messages
- Keep backward compatibility with existing import patterns
BREAKING CHANGE: Tools renamed from snake_case to lowercase
- read_file → read
- write_file → write
- list_directory → ls
- search_files → glob
- New tools: grep, delete
- Removed: edit_file, create_directory, directory_tree, move_file, get_file_info
* 🐛 fix: remove filesystem allowed directories restriction
* 🐛 fix: relax binary detection for text files
* ✨ feat: add edit tool with fuzzy matching to filesystem MCP server
- Add edit tool with 9 fallback replacers from opencode for robust
string replacement (SimpleReplacer, LineTrimmedReplacer,
BlockAnchorReplacer, WhitespaceNormalizedReplacer, etc.)
- Add Levenshtein distance algorithm for similarity matching
- Improve descriptions for all tools (read, write, glob, grep, ls, delete)
following opencode patterns for better LLM guidance
- Register edit tool in server and export from tools index
* ♻️ refactor: replace allowedDirectories with baseDir in filesystem MCP server
- Change server to use single baseDir (from WORKSPACE_ROOT env or userData/workspace default)
- Remove list_allowed_directories tool as restriction mechanism is removed
- Add ripgrep integration for faster grep searches with JS fallback
- Simplify validatePath() by removing allowlist checks
- Display paths relative to baseDir in tool outputs
* 📝 docs: standardize filesystem MCP server tool descriptions
- Unify description format to bullet-point style across all tools
- Add absolute path requirement to ls, glob, grep schemas and descriptions
- Update glob and grep to output absolute paths instead of relative paths
- Add missing error case documentation for edit tool (old_string === new_string)
- Standardize optional path parameter descriptions
* ♻️ refactor: use ripgrep for glob tool and extract shared utilities
- Extract shared ripgrep utilities (runRipgrep, getRipgrepAddonPath) to types.ts
- Rewrite glob tool to use `rg --files --glob` for reliable file matching
- Update grep tool to import shared ripgrep utilities
* 🐛 fix: handle ripgrep exit code 2 with valid results in glob tool
- Process ripgrep stdout when content exists, regardless of exit code
- Exit code 2 can indicate partial errors while still returning valid results
- Remove fallback directory listing (had buggy regex for root-level files)
- Update tool description to clarify patterns without "/" match at any depth
* 🔥 chore: remove filesystem.ts.backup file
Remove unnecessary backup file from mcpServers directory
* 🐛 fix: use correct default workspace path in filesystem MCP server
Change default baseDir from userData/workspace to userData/Data/Workspace
to match the app's data storage convention (Data/Files, Data/Notes, etc.)
Addresses PR #11937 review feedback.
* 🐛 fix: pass WORKSPACE_ROOT to FileSystemServer constructor
The envs object passed to createInMemoryMCPServer was not being used
for the filesystem server. Now WORKSPACE_ROOT is passed as a constructor
parameter, following the same pattern as other MCP servers.
* \feat: add link to documentation for MCP server configuration requirement
Wrap the configuration requirement tag in a link to the documentation for better user guidance on MCP server settings.
---------
Co-authored-by: kangfenmao <kangfenmao@qq.com>
Change the default knowledge base retrieval behavior from tool call to prompt injection mode.
This provides faster response times when knowledge base search is forced.
Intent recognition mode (tool call) is still available as an opt-in option.
- Remove toolChoiceMiddleware for forced knowledge base search
- Add prompt injection for knowledge base references in KnowledgeService
- Move transformMessagesAndFetch to ApiService, delete OrchestrateService
- Export getMessageContent from searchOrchestrationPlugin
- Add setCitationBlockId callback to citationCallbacks
- Default knowledgeRecognition to 'off' (prompt mode)
🤖 Generated with [Claude Code](https://claude.com/claude-code)
Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
Update toast component to use the standard sonner package instead of the cherrystudio fork. This simplifies dependencies and ensures better maintainability with the upstream package.
- Add custom toast component with support for info, success, warning, error, and loading states
- Implement action buttons, links, colored messages, and dismissable toasts
- Add TypeScript type definitions and Storybook documentation
- Export toast API for programmatic usage
- Add sonner package for toast notifications with custom icons and styling
- Add next-themes package for theme support
- Create new Toaster component with theme-aware styling
feat(assistants): merge import and subscribe popups, add export to manage
- Merge import and subscribe buttons into single unified popup
- Add export functionality to manage assistant presets
- Change delete mode to manage mode with both export and delete options
- Show import count in success message
- Default to manage mode when opening manage popup
- Fix unsubscribe button to clear URL properly
- Fix file import not working issue
🤖 Generated with [Claude Code](https://claude.com/claude-code)
Co-authored-by: Claude Opus 4.5 <noreply@anthropic.com>
* feat(topics): add topic manage mode for batch operations
- Add topic manage mode with batch delete and move operations
- Implement search functionality within manage mode with keyword matching
- Create reusable AssistantAvatar component for consistent icon display
- Add assistant icons to move-to dropdown menus
- Include selection badge with clear selection tooltip
- Add delete confirmation dialog with danger button styling
🤖 Generated with [Claude Code](https://claude.com/claude-code)
Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
* refactor(TopicManageMode): convert styled components to Tailwind CSS
- Replace styled-components with Tailwind CSS for ManagePanel, ManagePanelContent, ManageIconButton, and other UI elements.
- Update button styling to use Tailwind classes for improved consistency and maintainability.
- Enhance component structure with functional components and props for better reusability.
* style(Topics): update HeaderIconButton dimensions and border radius
- Increased dimensions of HeaderIconButton from 28px to 32px for improved visibility.
- Updated border radius to use a CSS variable for consistency with other UI elements.
---------
Co-authored-by: Claude Opus 4.5 <noreply@anthropic.com>
- Incremented the version number in the persisted reducer from 185 to 186.
- Added migration logic for version 186 to handle API server settings and OpenAI configuration updates, ensuring compatibility with existing user settings.
This change prepares the application for the new migration requirements and maintains backward compatibility.
* feat: add tool use mode setting to default assistant settings
- Add toolUseMode selector (prompt/function) to DefaultAssistantSettings
- Add dividers between model parameter sections for better UI
- Reduce slider margins for compact layout
- Add migration (v185) to reset toolUseMode to 'function' for existing users
🤖 Generated with [Claude Code](https://claude.com/claude-code)
Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
* fix: reset toolUseMode for all assistants during migration
- Update migration logic to reset toolUseMode to 'function' for all assistants with a 'prompt' setting.
- Ensure compatibility with function calling models by checking model type before resetting.
---------
Co-authored-by: Claude Opus 4.5 <noreply@anthropic.com>
Simplifies and centralizes header construction by merging defaultAppHeaders and extra_headers, and sets X-Api-Key for OpenAI providers. Removes redundant header assignment logic for improved maintainability.
- Removed the UI typecheck script from package.json to streamline type checking.
- Updated tsconfig.web.json to include the UI package for improved type checking coverage.
- Adjusted tsconfig.web.json to maintain formatting consistency.
- Enhanced tsconfig.json in the ui package by adding strict compiler options for better code quality.
- Modified the generate-icons script to remove an unused parameter, streamlining the function call.
- Introduced multiple new SVG icons including AddCategory, AiChat, Aicon27, AiEssentialsIconSet, AiPrompt, Brain, BrainCircuit, BrainCog, CodeAi, Emoji, Group, MessageAi1, MessageBalloonAi1, and Vector.
- Updated the icons index file to include the new icons for easier access.
- Enhanced the icons component structure for better maintainability.
* refactor: rename i18n commands for better consistency
- Rename `check:i18n` to `i18n:check`
- Rename `sync:i18n` to `i18n:sync`
- Rename `update:i18n` to `i18n:translate` (clearer purpose)
- Rename `auto:i18n` to `i18n:all` (runs check, sync, and translate)
- Update lint script to use new `i18n:check` command name
This follows the common naming convention of grouping related commands
under a namespace prefix (e.g., `test:main`, `test:renderer`).
🤖 Generated with [Claude Code](https://claude.com/claude-code)
Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
* refactor: update i18n command names and improve documentation
- Renamed i18n commands for consistency: `sync:i18n` to `i18n:sync`, `check:i18n` to `i18n:check`, and `auto:i18n` to `i18n:translate`.
- Updated relevant documentation and scripts to reflect new command names.
- Improved formatting and clarity in i18n-related guides and scripts.
This change enhances the clarity and usability of i18n commands across the project.
---------
Co-authored-by: Claude Opus 4.5 <noreply@anthropic.com>
* ✨ feat: add CDP browser MCP server
* ♻️ refactor: add navigation timeout for browser cdp
* 🐛 fix: reuse window for execute and add debugger logging
* ✨ feat: add show option and multiline execute for browser cdp
* ✨ feat: support multiple sessions for browser cdp
* ♻️ refactor: add LRU and idle cleanup for browser cdp sessions
* Refactor browser-cdp for readability and set Firefox UA
* 🐛 fix: type electron mock for cdp tests
* ♻️ refactor: rename browser_cdp MCP server to browser
Simplify the MCP server name from @cherry/browser-cdp to just browser
for cleaner tool naming in the MCP protocol.
* ✨ feat: add fetch tool to browser MCP server
Add a new `fetch` tool that uses the CDP-controlled browser to fetch URLs
and return content in various formats (html, txt, markdown, json).
Also ignore .conductor folder in biome and eslint configs.
* ♻️ refactor: split browser MCP server into modular folder structure
Reorganize browser.ts (525 lines) into browser/ folder with separate
files for better maintainability. Each tool now has its own file with
schema, definition, and handler.
* ♻️ refactor: use switch statement in browser server request handler
* ♻️ refactor: extract helpers and use handler registry pattern
- Add successResponse/errorResponse helpers in tools/utils.ts
- Add closeWindow helper to consolidate window cleanup logic
- Add ensureDebuggerAttached helper to consolidate debugger setup
- Add toolHandlers map for registry-based handler lookup
- Simplify server.ts to use dynamic handler dispatch
* 🐛 fix: improve browser MCP server robustness
- Add try-catch for JSON.parse in fetch() to handle invalid JSON gracefully
- Add Zod schema validation to reset tool for consistency with other tools
- Fix memory leak in open() by ensuring event listeners cleanup on timeout
- Add JSDoc comments for key methods and classes
* ♻️ refactor: rename browser MCP to @cherry/browser
Follow naming convention of other builtin MCP servers.
* 🌐 i18n: translate pending strings across 8 locales
Translate all "[to be translated]" markers including:
- CDP browser MCP server description (all 8 locales)
- "Extra High" reasoning chain length option (6 locales)
- Git Bash configuration strings (el-gr, ja-jp)
* ✨ feat: add ExaMCP free web search provider
Add a new web search provider that uses Exa's free MCP API endpoint,
requiring no API key. This provides users with a free alternative
to the existing Exa provider.
- Add 'exa-mcp' to WebSearchProviderIds
- Create ExaMcpProvider using JSON-RPC/SSE protocol
- Add provider config and migration for existing users
- Use same Exa logo in settings UI
* Add robust text chunk parser for ExaMcpProvider results
* feat(translate): add reasoning effort option to translate service
Add support for configuring reasoning effort level in translation requests. This allows better control over the translation quality and processing time based on model capabilities.
* test: add comprehensive tests for getModelSupportedReasoningEffort
* test(reasoning): update model test cases and comments
- Remove test case for 'gpt-4o-deep-research' as it needs to be an actual OpenAI model
- Add provider requirement comment for Grok 4 Fast recognition
- Simplify array assertions in test cases
- Add comment about Qwen models working well for name-based fallback
* docs(reasoning): add detailed jsdoc for getModelSupportedReasoningEffort
* refactor(openai): replace getThinkModelType with getModelSupportedReasoningEffort
Simplify reasoning effort validation by using getModelSupportedReasoningEffort
* refactor(models): rename getModelSupportedReasoningEffort to getModelSupportedReasoningEffortOptions
Update function name and all related references to better reflect its purpose of returning reasoning effort options
* fix(code-viewer): copy selected code without line numbers
* fix(context-menu): strip line numbers from code selection
* style(codeviewer): fix format
* fix: preserve indentation and format when copying mixed content (text + code blocks)
- Replace regex-based extraction with DOM structure-based approach
- Remove line number elements while preserving all other content
- Use TreeWalker to handle mixed content (text paragraphs + code blocks)
- Preserve indentation and newlines in code blocks
- Simplify CodeViewer.tsx by removing duplicate context menu logic
Fixes#11790
* style: remove unused comment
Co-authored-by: Copilot <175728472+Copilot@users.noreply.github.com>
* refactor: optimize TreeWalker performance
---------
Co-authored-by: Copilot <175728472+Copilot@users.noreply.github.com>
- Increased maxHeight of the modal body to 70vh for improved visibility.
- Changed LogList to extend Scrollbar for better scrolling experience.
- Updated LogItem background color to use a more appropriate variable for consistency.
- Add tool calling support in tooluse.ts
- Add reasoning support in reasoning.ts
- Add vision support in vision.ts
Doubao Seed Code models (doubao-seed-code-preview-251028 and future models)
now support function calling, deep thinking (enabled/disabled), and image understanding.
Updated the marginRight property in the Chat component to include the border width when the topic position is 'right' and topics are shown. This change enhances the layout by ensuring proper spacing in the UI.
* feat: support gpt 5.2
* feat: support param when set to 'none'
* chore version & simply type
* fix: comment
* fix: typecheck
* replace placeholder
* simplify func
* feat: add gpt-5.1-codex-max
* fix: correct token calculation in prompt tool use plugin
- Fix duplicate token accumulation in recursive stream handling
- Accumulate usage for finish-step without tool calls
- Filter out recursive stream's start/finish events (only one per conversation)
- Make accumulateUsage method public for reuse
🤖 Generated with [Claude Code](https://claude.com/claude-code)
Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
* refactor: simplify pipeRecursiveStream method in StreamEventManager
- Removed unnecessary context parameter from pipeRecursiveStream method
- Streamlined the invocation of pipeRecursiveStream in recursive call handling
This change enhances code clarity and reduces complexity in stream management.
---------
Co-authored-by: Claude Opus 4.5 <noreply@anthropic.com>
* fix: update Ollama provider options for Qwen model support
Pass the model to buildOllamaProviderOptions and enable 'think' option only for supported Qwen models. This improves reasoning capability handling for Ollama providers.
* fix: empty array
* feat: ollama oss
---------
Co-authored-by: suyao <sy20010504@gmail.com>
