flutter_ui_mcp_core

Shared UI components, dialogs, drawers, controllers, and services for Flutter MCP Suite packages: flutter_ui_mcp and flutter_genui_mcp.

✨ Features

  • Episodic Memory Archive & Retrieval (MemoryArchiveDialog) – Inspect, search, and navigate compressed episodic memory summaries and full-turn histories. Includes automatic background compression and built-in retrieve_memory tool integration in PlaygroundController.
  • Dynamic Model Pricing & Context Limits (ModelPricingService) – Real-time dynamic token pricing and context window resolution without hardcoded static tables (queries host /models catalogs, OpenRouter live API, DeepInfra, and local endpoints). Includes interactive banner with USD ($) and EUR (€) rates.
  • Autonomous Sub-Agent Delegation (SubAgentExecution) – Full observability and autonomous multi-turn tool execution loop for child sub-agents spawned via DelegateTaskTool. Shows task hierarchy, depth, live logs, status, and specialized execution cards inside ChatBubble.
  • Dynamic Skill Activation Dock (DynamicSkillDock) – Interactive chips dock allowing users and agents (ActivateSkillsTool/DeactivateSkillsTool) to activate/deactivate specialized skills at runtime without restarting the session.
  • Multimodal Chat & Lightbox Preview – Native support for typed ChatContentParts (ImagePart, DocumentPart, AudioPart). Provides inline previews, interactive full-screen image lightboxes, document chips, and audio pills.
  • Evaluation Benchmark Runner (EvalController, EvalBenchmarkDialog) – Run automated LLM and agent evaluations directly within Flutter applications. Features pass rate telemetry, latency tracking, custom test suites, and detailed grading rationales.
  • On-Device & Embedded AI (EmbeddedLlmAdapter) – Runs local on-device .gguf models in-process via llamadart (llama.cpp) across Android, iOS, Windows, macOS, Linux, and Web (WebGPU) with zero cloud latency and zero external keys. Ollama and local LLM servers connect seamlessly via standard OpenAI-compatible endpoints.
  • Multimodal Image Optimization (ImageOptimizer) – Automatically scales large camera photos down to optimal dimensions (max 768px) with native hardware decoders to prevent mobile NPU memory bottlenecks and out-of-memory errors.
  • PlaygroundController – Full lifecycle and state manager for agent conversations, tool execution, MCP server synchronization, and inspector tracing.
  • Workflow Automation & Execution – Define, save, load, and run multi-step agent workflows (WorkflowDef, WorkflowSaveDialog, WorkflowLoadDialog, WorkflowStep, sub-prompt pipelines).
  • Skills System & Skill Wizard – Comprehensive skill authoring and discovery (SkillsManagerDialog, SkillWizardDialog, SkillDetailsDialog, ActiveSkillBanner, SkillStorageAdapter, ZIP import/export).
  • LLM Configuration & Settings – LlmConfigForm, SettingsDrawer, SubPromptListEditor.
  • MCP Server & Tool Management – McpServerRegistryTab, EditMcpDialog, RemoteMcpDialog, RegisteredToolsDialog, ServerToolsDialog.
  • On-Device Embedded LLMs – EmbeddedModelManager, EmbeddedModelPickerWidget, HfDiscoverDialog, AddGgufDialog.
  • Inspector Panel – AgentInspector side pane for real-time observability of tool executions, raw JSON payloads, and system prompts.
  • Localization – Full English & German strings via McpLocalizations.

🎥 Demo

Watch shared components in action executing tasks with local tools and embedded models:

Flutter UI MCP Core Demo

📦 Installation

Add to your pubspec.yaml:

dependencies:
  flutter_ui_mcp_core: ^1.3.0
  dart_mcp_core: ^1.3.0

📂 Examples

Libraries

flutter_ui_mcp_core
playground_controller