CodeGraph CLI MCP Server

# CodeGraph Configuration Example # Copy this file to .env and update the values for your environment # ============================================================================ # CodeGraph Core Configuration (Simplified Setup) # ============================================================================ # Minimal Setup - Auto-detect embedding provider (ONNX, Ollama, or OpenAI) CODEGRAPH_EMBEDDING_PROVIDER=auto # That's it for basic usage! CodeGraph will auto-detect everything else. # Uncomment and customize the settings below if you need more control. # ============================================================================ # Storage and Installation Configuration # ============================================================================ # Storage path for CodeGraph data (indexes, cache, etc.) # CODEGRAPH_STORAGE=/path/to/storage # CODEGRAPH_STORAGE_PATH=/path/to/storage # Alternative name # Installation directory for binaries # CODEGRAPH_INSTALL_DIR=$HOME/.local/bin # Project identification (for multi-project setups) # CODEGRAPH_PROJECT_ID=my-project # CODEGRAPH_ORGANIZATION_ID=my-org # CODEGRAPH_REPOSITORY_URL=https://github.com/user/repo # CODEGRAPH_DOMAIN=example.com # ============================================================================ # Embedding Provider Configuration # ============================================================================ # Provider options: "auto", "onnx", "ollama", "openai", "jina", or "lmstudio" # CODEGRAPH_EMBEDDING_PROVIDER=auto # ONNX: Specify model path (or leave empty for auto-detection from HuggingFace cache) # CODEGRAPH_LOCAL_MODEL=/path/to/your/onnx/model # Ollama: Specify embedding model name # CODEGRAPH_EMBEDDING_MODEL=all-minilm:latest # CODEGRAPH_OLLAMA_URL=http://localhost:11434 # LM Studio: Best for MLX + Flash Attention 2 (recommended on macOS) # Default: jina-embeddings-v4 (2048 or 1024 dimensions) # CODEGRAPH_EMBEDDING_PROVIDER=lmstudio # CODEGRAPH_EMBEDDING_MODEL=jinaai/jina-embeddings-v4 # CODEGRAPH_LMSTUDIO_URL=http://localhost:1234 # CODEGRAPH_EMBEDDING_DIMENSION=2048 # Batch size for embedding generation (applies to all providers) # CODEGRAPH_EMBEDDING_BATCH_SIZE=32 # Default: 32, valid range: 1-4096 # OpenAI: Model name (API key configured below in Security section) # CODEGRAPH_EMBEDDING_MODEL=text-embedding-3-small # Jina AI: Cloud embeddings with reranking (requires JINA_API_KEY) # CODEGRAPH_EMBEDDING_PROVIDER=jina # JINA_EMBEDDING_MODEL=jina-embeddings-v4 # JINA_EMBEDDINGS_MODEL=jina-embeddings-v4 # Alternative name # JINA_EMBEDDING_DIMENSION=2048 (supports 1024, 512, 256 but only 1024 has pre-defined column and HNSW index in surrealdb schema) # JINA_API_KEY=your-jina-api-key-here # JINA_MAX_TEXTS=512 # Leverage Jina API Batch functionality max 512 documents with 8192 tokens each remember to set --max-concurrent 1 when indexing # JINA_MAX_TOKENS=7000 # JINA_API_BASE=https://api.jina.ai/v1 # JINA_API_TASK=code.passage # used when embedding data, code.query is used when searching data # JINA_TASK=code.passage # Alternative name # JINA_LATE_CHUNKING=false # JINA_TRUNCATE=true # truncate texts and embeddings if over limit # JINA_REQUEST_DELAY_MS=600 # small delay not to throttle the API when batching # Jina Reranking Configuration # CODEGRAPH_RERANKING_PROVIDER=jina # CODEGRAPH_RERANK_CANDIDATES=512 # Leverage Jina API Batch functionality also here # JINA_ENABLE_RERANKING=true # JINA_RERANKING_ENABLED=true # Alternative name # JINA_RERANKING_MODEL=jina-reranker-v3 # JINA_RERANKING_TOP_N=10 # Jina Relationship Embeddings Configuration # JINA_REL_BATCH_SIZE=50 # JINA_REL_MAX_TEXTS=50 # Jina Batching for Large Indexing Operations # CODEGRAPH_JINA_BATCH_SIZE=2000 # CODEGRAPH_JINA_BATCH_MINUTES=9.0 # used by estimate command, this is how long it took on average to index the codegraph codebase nodes with Jina # ============================================================================ # Dual-Mode Search Configuration # ============================================================================ # CodeGraph supports two search modes based on CODEGRAPH_EMBEDDING_PROVIDER: # # Local Mode (FAISS + local/ollama embeddings) # --------------------------------------------- # - Uses FAISS for in-memory vector search # - Embeddings: ONNX, Ollama, or LM Studio # - Best for: Desktop development, privacy-focused setups # - Requires: Build with --features faiss # Example: # CODEGRAPH_EMBEDDING_PROVIDER=local # or ollama or lmstudio # # Cloud Mode (SurrealDB HNSW + Jina embeddings + reranking) # ---------------------------------------------------------- # - Uses SurrealDB HNSW indexes for scalable vector search # - Embeddings: Jina AI (Variable Matryosha dimensions - check what the model outputs and adjust schema/codegraph.surql HNSW vector dims) # - Supported Jina AI embedding model is f.ex. jina-embeddings-v4 # - Reranking: Jina reranker-v3 for improved relevance # - Best for: Cloud deployments, multi-user systems, scalability # - Requires: SurrealDB instance, Jina API key # Example: # CODEGRAPH_EMBEDDING_PROVIDER=jina # JINA_API_KEY=your-jina-api-key-here # Cloud features toggle # CODEGRAPH_CLOUD_ENABLED=false # ============================================================================ # SurrealDB Configuration (required for cloud mode) # ============================================================================ # SurrealDB connection (local or cloud) # CODEGRAPH_SURREALDB_URL=ws://localhost:3004 # SURREALDB_URL=ws://localhost:3004 # Alternative name # CODEGRAPH_SURREALDB_NAMESPACE=codegraph # SURREALDB_NAMESPACE=codegraph # Alternative name # CODEGRAPH_SURREALDB_DATABASE=main # SURREALDB_DATABASE=main # Alternative name # CODEGRAPH_SURREALDB_USERNAME=root # SURREALDB_USERNAME=root # Alternative name # CODEGRAPH_SURREALDB_PASSWORD=root # SURREALDB_PASSWORD=root # Alternative name # # Important: HNSW index dimension must match embedding provider # - Jina: Variable Matryoska dimensions depending on model 2048, 1024, 512, 256 # - OpenAI: Small 1536 dimensions, Large 3072 dimensions # - Local ONNX: typically 384 qdrant/all-mini-llm-onnx # - Local Ollama: qwen3-embedder:0.6b-8b 1024, 2048, 4096 # - 384 (all-mini-llm:latest) # - 1024 (qwen3-embedding:0.6b) # - 1536 (text-embedding-3-small) # - 2048 (qwen3-embedding:4b, jina-embeddings-v4) # - 3072 (text-embedding-3-large) # - 4096 (qwen3-embedding:8b) # The following embedding fields and HNSW indexes have been built into the surrealdb schema and codegraph codebase # For pure speed use onnx or ollama all-mini-llm # For considerably better retriaval switch to Ollama and qwen3-embedding:0.6b # Scale qwen3-embedding model per criticality of accuracy on Ollma # For enhancing local accuracy enable CODEGRAPH_RERANKING_PROVIDER=lmstudio and f.ex. CODEGRAPH_RERANKING_MODEL=qwen-reranker-3:0.6b # For true SOTA use jina provider and jina-embeddings-v4 with the jina-reranker-v3 - takes longer to index but works better # Hierarchical config system also supports CODEGRAPH__* prefix for nested config # Example: CODEGRAPH__DATABASE__BACKEND=surrealdb # Example: CODEGRAPH__DATABASE__SURREALDB__CONNECTION=ws://localhost:8000 # Example: CODEGRAPH__DATABASE__SURREALDB__PASSWORD=your_password # Example: CODEGRAPH__SERVER__PORT=8080 # Example: CODEGRAPH__LOGGING__LEVEL=debug # ============================================================================ # LLM Configuration (for local insights generation) # ============================================================================ # Leave empty to use context-only mode (fastest, recommended for agents like Claude/GPT-4) # Set to enable local LLM insights generation # LM Studio with DeepSeek Coder v2 Lite Instruct (or what ever fits in your vGPU memory) # Supported LLM provider options: "auto", "onnx", "lmstudio", "openai", "claude" or "ollama" # Superior MLX support, Flash Attention 2, KV-cache and Distillation model support on macOS # CODEGRAPH_LLM_PROVIDER=lmstudio # LLM_PROVIDER=lmstudio # Alternative name # CODEGRAPH_MODEL=lmstudio-community/DeepSeek-Coder-V2-Lite-Instruct-GGUF/DeepSeek-Coder-V2-Lite-Instruct-Q4_K_M.gguf # CODEGRAPH_LMSTUDIO_URL=http://localhost:1234 # CODEGRAPH_CONTEXT_WINDOW=32000 # CODEGRAPH_TEMPERATURE=0.1 # Reasoning effort for complex queries (affects token usage) # CODEGRAPH_REASONING_EFFORT=medium # Ollama (alternative) # LLM model (e.g., "qwen2.5-coder:14b", "Kimi-K2-Instruct") # CODEGRAPH_MODEL=qwen2.5-coder:14b # CODEGRAPH_OLLAMA_URL=http://localhost:11434 # CODEGRAPH_CONTEXT_WINDOW=252000 # Max for Ollama depends on the model though # Anthropic (cloud - 200K/1M tokens) # CODEGRAPH_LLM_PROVIDER=anthropic # CODEGRAPH_MODEL=sonnet[1m] # ANTHROPIC_API_KEY=sk-ant-your-key-here # CODEGRAPH_CONTEXT_WINDOW=1000000 # 200K/1M tokens # OpenAI (cloud - 200K/400K tokens) # CODEGRAPH_LLM_PROVIDER=openai # CODEGRAPH_MODEL=gpt-5-codex-mini # OPENAI_API_KEY=sk-your-key-here # OPENAI_ORG_ID=your_fabulous_org # CODEGRAPH_CONTEXT_WINDOW=200000 # 400K tokens # CODEGRAPH_REASONING_BUDGET=medium # xAI (cloud - 252K/2M context window, $0.50-$1.50/M tokens!) # CODEGRAPH_LLM_PROVIDER=openai # OpenAI Responses API compatible providers work # CODEGRAPH_MODEL=grok-4-fast # or grok-code-fast-1 252K tokens the price is same # XAI_API_KEY=xai-your-key-here # CODEGRAPH_CONTEXT_WINDOW=2000000 # 2M tokens! # CODEGRAPH_REASONING_BUDGET=high # MCP-server code insights agent max output tokens - uses the CODEGRAPH_MODEL # MCP_CODE_AGENT_MAX_OUTPUT_TOKENS=52000 # ============================================================================ # Performance & Caching Configuration # ============================================================================ # Performance mode (affects optimization settings) # CODEGRAPH_PERFORMANCE_MODE=balanced # Options: high_speed, balanced, low_memory # Cache configuration # CODEGRAPH_CACHE_SIZE=1000 # Maximum cache entries # CODEGRAPH_CACHE_TTL=1800 # Cache TTL in seconds # CODEGRAPH_ENABLE_CACHE=true # Symbol indexing batch sizes # CODEGRAPH_SYMBOL_BATCH_SIZE=500 # Batch size for symbol processing # CODEGRAPH_SYMBOL_MAX_CONCURRENT=4 # Max concurrent symbol processing # CODEGRAPH_SYMBOL_DB_BATCH_SIZE=1000 # Batch size for database writes # ============================================================================ # Qwen-Coder-2.5-128K Integration (Legacy) # ============================================================================ # Qwen model configuration # CODEGRAPH_ENABLE_QWEN=true # CODEGRAPH_QWEN_MAX_TOKENS=128000 # Max completion tokens (0 disables limit) # CODEGRAPH_QWEN_TIMEOUT_SECS=180 # Request timeout before fallback (0 disables) # CODEGRAPH_QWEN_CONNECT_TIMEOUT_MS=5000 # Connection timeout to Ollama # ============================================================================ # Server Configuration # ============================================================================ # Server host and port (for HTTP/REST API) # CODEGRAPH_HOST=127.0.0.1 # CODEGRAPH_PORT=3000 # Logging # ------- # Log level: trace, debug, info, warn, error # Use "warn" during indexing for clean TUI output (recommended) # Use "info" for development/debugging RUST_LOG=warn # ============================================================================ # Testing & Development Configuration # ============================================================================ # Path to codegraph binary (for testing) # CODEGRAPH_BIN=./target/release/codegraph # CODEGRAPH_BIN=./target/debug/codegraph # Command to run codegraph (alternative to binary path) # CODEGRAPH_CMD="cargo run -p codegraph-mcp --bin codegraph --" # ============================================================================ # Security Configuration (for production deployments) # ============================================================================ # JWT Authentication JWT_SECRET=replace_with_secure_random_secret_minimum_32_characters_long JWT_EXPIRY_HOURS=24 # API Key Configuration API_KEY_PREFIX=cgk # Server Configuration HOST=127.0.0.1 PORT=8080 ENVIRONMENT=development # TLS/HTTPS Configuration (for production) # TLS_CERT_PATH=/path/to/certificate.pem # TLS_KEY_PATH=/path/to/private-key.pem # REQUIRE_TLS=true # Database Configuration # DATABASE_URL=postgresql://user:password@localhost/codegraph # REDIS_URL=redis://localhost:6379 # Rate Limiting RATE_LIMIT_ANONYMOUS=60 RATE_LIMIT_USER=1000 RATE_LIMIT_PREMIUM=5000 RATE_LIMIT_ADMIN=10000 # Security Settings MAX_REQUEST_SIZE=10485760 # 10MB SESSION_TIMEOUT_HOURS=24 PASSWORD_MIN_LENGTH=12 # Logging (see RUST_LOG above for CodeGraph core logging) # LOG_LEVEL=info # For application-level logging SECURITY_LOG_LEVEL=warn LOG_FORMAT=json # Monitoring METRICS_ENABLED=true PROMETHEUS_PORT=9090 # External Services # SENTRY_DSN=https://your-sentry-dsn # ANALYTICS_KEY=your_analytics_key # Development/Testing Only DEV_MODE=true DISABLE_AUTH=false # Never set to true in production! ENABLE_DEBUG_ENDPOINTS=false # ============================================================================ # HTTP Server Configuration (when using --transport http) # ============================================================================ # Host address to bind HTTP server (default: 127.0.0.1) # Use 0.0.0.0 to allow external connections CODEGRAPH_HTTP_HOST=127.0.0.1 # Port for HTTP server (default: 3000) CODEGRAPH_HTTP_PORT=3000 # SSE keep-alive interval in seconds (default: 15) # Prevents proxy timeouts for long-running agentic operations CODEGRAPH_HTTP_KEEP_ALIVE=15