VibeScout
Enables semantic code search and AI-powered chat for local codebases, with support for multiple AI providers and a plugin system.
README
VibeScout: Local Code Intelligence & MCP Server
A high-performance Model Context Protocol (MCP) server and Web Dashboard for local semantic code search and AI-powered assistance. VibeScout transforms your codebase into a searchable, chat-ready knowledge base using local or cloud-based AI providers.
🚀 Features
Core Capabilities
- Web Dashboard: A modern React-based UI for visual searching, chatting with your code, and managing your knowledge base.
- Semantic Code Search: Fast vector-based search with BGE reranking for extreme technical accuracy.
- Visual Dependency Graph: Explore architecture visually with Symbol Intelligence panel and Circular Dependency Detection.
- AI-Powered Chat: Persistent context chat with RAG-powered code understanding.
- Real-time File Watching: Automatic re-indexing when files change (chokidar-based).
Search & Filtering
- Category Pre-Filtering: Database-level filtering by Code/Docs for maximum performance (default: Code-only).
- Token Count Preview: Preview search results before consuming tokens with two-phase search (MCP only).
- Git Metadata: Enhanced search with author, date, churn level, and commit history.
- Dependencies Display: View imports/exports for each search result with source tracking.
- Search Persistence: Results persist across tab switches for seamless workflow.
- Framework Detection: Auto-detects Next.js, React Router, Vue, Angular, and 10+ frameworks.
Plugin System
- Modular Architecture: Built-in, npm, and local plugin support with override detection.
- Versioned Structure:
/src/plugins/<name>/<version>/for built-in plugins. - Compatibility Checking: Auto-disables incompatible plugins with clear warnings.
- Runtime Toggle: Enable/disable plugins via UI or config file.
- Sandboxed Execution: Isolated worker threads for safe plugin operation.
Advanced Features
- Two-Phase Search: Preview search metadata before consuming tokens (MCP only).
- AI Smart Questions: "Generate Best Question" analyzes code to suggest optimal chat starting points.
- Adaptive Concurrency: Automatically detects provider rate limits and dynamically scales request rates.
- Real-time Activity (SSE): Instant system event streaming via Server-Sent Events.
- AI Inspector (Debug): Chrome DevTools-style panel to inspect prompts, payloads, and responses.
- IDE Integration: "Open in Editor" with line-level navigation for VS Code and Cursor.
- Performance Profiling: Built-in profiling with Chrome DevTools flame graphs for optimizing indexing and search performance.
Technical Stack
- Hono-Powered Backend: Ultra-low latency with standard Web Fetch API support.
- Multi-Language Support: TypeScript/JS, Python, Go, Java, Kotlin, Dart, Markdown, JSON, TOML, XML.
- Hybrid Storage: Local LanceDB or Cloudflare Vectorize for cloud embeddings.
- AI Reranker: Optional second-pass re-sorting with local models.
- Fully Offline Mode: Strict local-only operation, no remote model downloads.
🤖 Supported Providers
| Provider | Type | Description |
|---|---|---|
| Local | Built-in | Transformers.js (BGE, MiniLM). High-speed, zero config. |
| Ollama | Local API | Full support for local models like Llama 3, Mistral, and Nomic. |
| Z.AI | Cloud | BigModel.cn integration. Supports GLM-4 and dedicated Coding Plans. |
| Bedrock | Cloud | AWS Bedrock (Claude 3, Titan). Supports AWS Profiles. |
| Gemini | Cloud | Google Generative AI (1.5 Flash/Pro) and specialized embeddings. |
| OpenAI | Cloud | Standard OpenAI API or compatible (DeepSeek, Groq, etc). |
| Cloudflare | Cloud | Workers AI & Vectorize integration for cloud-native indexing. |
| LM Studio | Local API | OpenAI-compatible local server preset. |
🔌 Plugin System
VibeScout features a powerful plugin system for extending functionality:
Plugin Sources
- Built-in Plugins: Shipped with VibeScout (
/src/plugins/<name>/<version>/) - npm Packages: Install via
vibescout plugin install <name> - Local Plugins: Place in
~/.vibescout/plugins/
Plugin Capabilities
- Extractors: Custom code extraction strategies (e.g., framework-specific metadata)
- Providers: Custom embedding/summarizer providers
- Commands: CLI commands for framework-specific operations
Plugin Management
# List plugins
vibescout plugin list
# Install from npm
vibescout plugin install vibescout-plugin-nextjs
# Install specific version
vibescout plugin install vibescout-plugin-nextjs@beta
# Uninstall
vibescout plugin uninstall vibescout-plugin-nextjs
# Enable/Disable (via Web UI or config.json)
# Plugins are stored in ~/.vibescout/config.json
Plugin Development
See /docs/ directory for:
plugin-guide.md- Getting startedplugin-api.md- API referenceplugin-architecture.md- Design patternsplugin-example.md- Complete exampletwo-phase-search.md- Two-phase search with token count previewprofiling-guide.md- Performance profiling and optimization
🛠Installation
Global Installation
npm install -g @sevenseconds/vibescout
💻 CLI Usage
Web UI
Launch the interactive dashboard:
vibescout ui
Advanced Logging
Control terminal output verbosity:
# Default is INFO (concise)
vibescout ui --log-level warn
# Full debug output (alias for --log-level debug)
vibescout ui --verbose
Indexing & Search
# Index a project manually
vibescout index ./my-app "My Project"
# Semantic search via terminal
vibescout search "how does the auth flow work?"
# Reset the database (clear all data)
vibescout reset
# Or via npm
npm run reset-db
Two-Phase Search (MCP Only)
When using VibeScout with Claude Desktop, Cursor, or other MCP clients, you can preview search results before consuming tokens:
Phase 1: Preview Token Count
{
"name": "search_code",
"arguments": {
"query": "authentication flow",
"limit": 20,
"previewOnly": true
}
}
This returns metadata without actual code:
- Result count
- Total tokens (from stored counts during indexing)
- Average relevance score
- Recommendation on whether to proceed
Phase 2: Fetch Results (if tokens are acceptable)
{
"name": "search_code",
"arguments": {
"query": "authentication flow",
"limit": 10,
"previewOnly": false
}
}
Benefits:
- Avoid unexpected token consumption
- Adjust
limitparameter based on preview - Use more specific filters if token count is high
- Backward compatible: omit
previewOnlyfor existing behavior
Performance Profiling
# Profile with 100% sampling (most detailed)
vibescout --profile index ./my-app "My Project"
# Profile with 10% sampling (lower overhead)
vibescout --profile --profile-sampling 0.1 search "authentication"
# Dedicated profiling command
vibescout profile index --folder ./my-app --sampling 1.0
# View traces in chrome://tracing
# Traces saved to ~/.vibescout/profiles/
Profiling Features:
- Zero overhead when disabled (default)
- Chrome DevTools-compatible flame graphs
- Configurable sampling rates (0.0-1.0) to reduce overhead
- Category-based sampling (indexing, search, embedding, database)
- Web UI dashboard at http://localhost:3000/performance
See docs/profiling-guide.md for detailed documentation.
🔒 Offline Mode
To use VibeScout in a restricted environment without internet access:
- Download Models: In an online environment, let VibeScout download the models first, or download them manually from Hugging Face.
- Enable Offline Mode: Toggle Offline Mode in Settings or run with the
--offlineflag. - Local Path: Use
--models-path <path>if your models are stored in a non-standard directory.
When enabled, VibeScout sets allowRemoteModels: false and disables all attempts to connect to the Hugging Face Hub.
🔌 Client Integration
Claude Desktop / Gemini CLI
Add VibeScout to your configuration:
{
"mcpServers": {
"vibescout": {
"command": "vibescout",
"args": ["--mcp", "stdio"]
}
}
}
📄 License
MIT License. See LICENSE for details.
Recommended Servers
playwright-mcp
A Model Context Protocol server that enables LLMs to interact with web pages through structured accessibility snapshots without requiring vision models or screenshots.
Magic Component Platform (MCP)
An AI-powered tool that generates modern UI components from natural language descriptions, integrating with popular IDEs to streamline UI development workflow.
Audiense Insights MCP Server
Enables interaction with Audiense Insights accounts via the Model Context Protocol, facilitating the extraction and analysis of marketing insights and audience data including demographics, behavior, and influencer engagement.
VeyraX MCP
Single MCP tool to connect all your favorite tools: Gmail, Calendar and 40 more.
graphlit-mcp-server
The Model Context Protocol (MCP) Server enables integration between MCP clients and the Graphlit service. Ingest anything from Slack to Gmail to podcast feeds, in addition to web crawling, into a Graphlit project - and then retrieve relevant contents from the MCP client.
Kagi MCP Server
An MCP server that integrates Kagi search capabilities with Claude AI, enabling Claude to perform real-time web searches when answering questions that require up-to-date information.
E2B
Using MCP to run code via e2b.
Neon Database
MCP server for interacting with Neon Management API and databases
Exa Search
A Model Context Protocol (MCP) server lets AI assistants like Claude use the Exa AI Search API for web searches. This setup allows AI models to get real-time web information in a safe and controlled way.
Qdrant Server
This repository is an example of how to create a MCP server for Qdrant, a vector search engine.