Advanced codebase indexing and semantic search MCP server
npm install @wildcard-ai/deepcodexA comprehensive Model Context Protocol (MCP) server by Wildcard Corporation that provides advanced codebase indexing and semantic search with slash commands and natural language interface for Claude Code.



/index, /search, /status, /clear, /context, /deps, /helpbash
npm install
npm run build
`$3
#### Required APIs:
- Jina AI: Get your API key from Jina AI - For embeddings and reranking
- Turbopuffer: Get your API key from Turbopuffer - For vector storage
$3
`bash
Install via npm
npm install -g @wildcard-corp/intelligent-context-mcpAdd to Claude Code
claude mcp add intelligent-context \\
-e JINA_API_KEY=your-jina-key \\
-e TURBOPUFFER_API_KEY=your-turbopuffer-key \\
-- npx @wildcard-corp/intelligent-context-mcp
`š® Usage
$3
Execute commands using slash syntax:
`bash
Index your codebase with intelligent chunking
/index /path/to/your/projectSearch with semantic understanding
/search authentication implementation
/search user registration flow
/search database connection setupCheck indexing status
/status
/status /specific/project/pathGet focused context for specific files or symbols
/context src/auth.js --with-deps
/context UserService --window=10Analyze dependencies
/deps src/user.js --reverse
/deps AuthController --graphClear index data
/clear --confirm
/clear /path/to/project --confirmGet help
/help
/help search
`$3
Use conversational queries:
`bash
"Find all authentication functions"
"Show me the user registration flow"
"Index my codebase at /path/to/project"
"What's the status of my index?"
"Get context for the login function"
`$3
Direct tool calls for programmatic access:
-
execute_slash_command - Execute any slash command
- natural_language_query - Process natural language queries
- index_codebase_intelligent - Direct indexing (legacy)
- search_with_intelligence - Direct search (legacy)āļø Configuration
$3
| Variable | Required | Description |
|----------|----------|-------------|
|
JINA_API_KEY | ā
| Your Jina AI API key for embeddings |
| TURBOPUFFER_API_KEY | ā
| Your Turbopuffer API key for vector storage |
| LOG_LEVEL | ā | Logging level: debug, info, warn, error (default: info) |
| CODEX_CONTEXT_DATA_DIR | ā | Data storage directory (default: ~/.codex-context) |$3
- Jina AI Embeddings: Uses
jina-embeddings-v3 model with 1024 dimensions
- Turbopuffer Storage: Vector storage with cosine distance similarity
- Tree-sitter Parsing: AST parsing for accurate symbol extractionšļø Architecture
$3
`mermaid
graph TD
A[Claude Code CLI] --> B[Enhanced MCP Interface]
B --> C{Interface Type}
C -->|Slash Commands| D[Command Parser]
C -->|Natural Language| E[NL Interpreter]
C -->|Direct Tools| F[Tool Handler]
D --> G[Integration Bridge]
E --> G
F --> G
G --> H[Core Components]
H --> I[IndexingOrchestrator]
H --> J[SemanticSearchEngine]
H --> K[TreeSitterSymbolExtractor]
I --> L[Real APIs]
J --> L
K --> L
`$3
1. Enhanced MCP Interface (
src/enhanced-mcp.ts)
- Primary interface with slash commands and natural language
- MCP protocol handling and state management
- Command registry with extensible architecture2. Integration Bridge (
src/standalone-mcp-integration.ts)
- Connects interface layer with core components
- Real API integration (Jina AI + Turbopuffer)
- Data format conversion and error handling3. Core Intelligence Components (
src/core/)
- IndexingOrchestrator: Advanced codebase indexing
- SemanticSearchEngine: Multi-strategy intelligent search
- TreeSitterSymbolExtractor: AST-based symbol extraction
- ContentFilterProvider: Quality filtering and noise reduction
- IncrementalIndexer: Efficient change detection and updatesš Performance & Quality
$3
- Symbol-boundary chunking: No arbitrary line splits
- Content filtering: Excludes ~40% of noise (tests, configs, generated code)
- Incremental updates: Only processes changed files
- Batch processing: Optimized API calls and vector uploads$3
- Multi-stage ranking: Vector similarity + optional reranking
- Dependency expansion: Finds related code across file boundaries
- Context windows: Configurable context around matches
- Symbol awareness: Understands code structure and relationshipsš Examples
$3
`bash
/index /path/to/react-app --force
ā
Successfully indexed 1,247 files into 3,821 intelligent chunks
š Ready for intelligent search with
/search
`$3
`bash
/search user authentication login
š Found 15 results (234ms):
#
src/auth/AuthService.ts:45-67 (0.923)
`typescript
async authenticateUser(credentials: LoginCredentials): Promise {
const user = await this.userRepository.findByEmail(credentials.email);
if (!user || !await this.verifyPassword(credentials.password, user.hashedPassword)) {
throw new UnauthorizedError('Invalid credentials');
}
return this.generateAuthTokens(user);
}
`
`$3
`bash
/context src/auth/AuthService.ts --with-deps
š Context for AuthService with dependencies:
- Depends on: UserRepository, TokenService, PasswordHasher
- Used by: LoginController, SignupController, AuthMiddleware
- Related symbols: authenticateUser, verifyPassword, generateTokens
`š Benefits
$3
- Intuitive interface with familiar slash commands
- Natural language queries for non-technical stakeholders
- Comprehensive context with dependency awareness
- Fast search with semantic understanding$3
- Consistent indexing with quality filtering
- Cross-file understanding via dependency analysis
- Incremental updates for active development
- Production ready with robust error handlingš Advanced Usage
$3
`bash
Semantic search for concepts
/search "how does authentication work" --type=semanticStructural search for patterns
/search "function.login.password" --type=structuralHybrid search combining both
/search authentication --type=hybrid
`$3
`bash
Find all files that depend on AuthService
/deps AuthService --reverseShow dependency graph for authentication module
/deps src/auth/ --graph --depth=3
`$3
`bash
Get 20 lines of context around matches
/search database connection --window=20Focus on specific symbol types
/context UserService --focus=functions --with-deps
`š ļø Development
$3
`
āāā src/ # Source code
ā āāā core/ # Core indexing and search logic
ā āāā services/ # Search and utility services
ā āāā utils/ # Utilities (Logger, FileUtils, etc.)
ā āāā types/ # TypeScript type definitions
āāā tests/ # Essential test files
āāā scripts/ # Utility scripts
āāā dist/ # Compiled JavaScript output
`$3
`bash
Set environment variables
export JINA_API_KEY="your_jina_api_key"
export TURBOPUFFER_API_KEY="your_turbopuffer_key"Test semantic chunking quality
node tests/final-chunking-validation.mjsTest search integration end-to-end
node tests/test-improved-search-quality.mjsTest search result quality directly
node tests/test-search-results-direct.mjsTest MCP server functionality
node tests/test-mcp-tools-directly.mjs
``Unlike simple text search tools, this MCP provides:
1. True code understanding via AST parsing and symbol extraction
2. Quality-focused indexing that filters out noise automatically
3. Dependency-aware search that finds related code across files
4. Multiple interfaces to suit different user preferences
5. Production deployment with real API integrations
6. Incremental efficiency for large, active codebases
Perfect for teams using Claude Code who need intelligent codebase exploration and context-aware development assistance.
---
Ready to enhance your codebase exploration with Claude Code! š