lxDIG — Dynamic Intelligence Graph Architecture

Overview

lxDIG MCP is a production Dynamic Intelligence Graph (DIG) server — the next evolution beyond static RAG and batch GraphRAG — that turns any repository into a queryable, incrementally-updated graph + retrieval system. It exposes 33 MCP tools across code intelligence, architecture validation, test selection, agent memory, and multi-agent coordination.

Server Implementation

Two server entry points exist — both production-ready:

File	Transport	Use case
`src/server.ts`	MCP HTTP (Streamable HTTP)	Production — multi-client, multi-session
`src/index.ts`	stdio (legacy)	Backward compat only

Recommended: src/server.ts via npm run start:http for agent fleets; stdio for local editor use via npm run start.

Key Architectural Properties

Session-scoped workspace context: each MCP session has its own project context (set via graph_set_workspace); no shared global state
Async graph rebuild: graph_rebuild returns status: QUEUED; poll via graph_health / graph_query until ready
Hybrid retrieval: natural graph_query uses vector + BM25/lexical + graph expansion with Reciprocal Rank Fusion (RRF)
Response shaping: three profiles (compact, balanced, debug) with token budget enforcement

Data Planes

┌──────────────────────────────────────────────────────────────┐
│  MCP Client (VS Code / agent / curl)                         │
└──────────────────┬───────────────────────────────────────────┘
                   │ MCP HTTP (Streamable HTTP)
┌──────────────────▼───────────────────────────────────────────┐
│  src/server.ts  — McpServer (MCP SDK)                        │
│  39 registered tools → ToolHandlers                          │
└────────┬──────────────────────────────────────────┬──────────┘
         │                                          │
┌────────▼──────────┐                  ┌────────────▼──────────┐
│  Memgraph (MAGE)  │                  │  Qdrant               │
│  graph plane      │                  │  vector plane         │
│  FILE/FUNCTION/   │                  │  embeddings per       │
│  CLASS/IMPORT     │                  │  function/class/file  │
│  + episodic nodes │                  │                       │
│  SCIP IDs on all  │                  │                       │
│  code nodes       │                  │                       │
└───────────────────┘                  └───────────────────────┘

Graph Schema

Node labels and key properties:

Label	Key properties
`FILE`	`path`, `language`, `projectId`, `scipId`, `checksum`
`FUNCTION`	`name`, `startLine`, `endLine`, `language`, `projectId`, `scipId`, `scopePath`
`CLASS`	`name`, `startLine`, `endLine`, `language`, `projectId`, `scipId`
`IMPORT`	`source`, `symbols[]`
`EPISODE`	`type`, `content`, `agentId`, `taskId`, `timestamp`, `projectId`
`CLAIM`	`agentId`, `taskName`, `claimedAt`, `projectId`

SCIP IDs (scipId field added to all code nodes):

File: src/tools/tool-handlers.ts
Function: src/tools/tool-handlers.ts::callTool()
Method: src/tools/tool-handlers.ts::ToolHandlers#arch_suggest()
Class: src/tools/tool-handlers.ts::ToolHandlers#

Relationships: CONTAINS, IMPORTS, CALLS, EXTENDS, IMPLEMENTS

Parser Architecture

Parsing is handled in src/graph/orchestrator.ts which dispatches to the appropriate parser per file extension.

Parser registry

Language	Extensions	Parser (default)	Parser (tree-sitter, `LXDIG_USE_TREE_SITTER=true`)
TypeScript	`.ts`	regex (typescript-parser.ts)	`TreeSitterTypeScriptParser`
TSX	`.tsx`	regex fallback	`TreeSitterTSXParser`
JavaScript	`.js`, `.mjs`, `.cjs`	FILE node only	`TreeSitterJavaScriptParser`
JSX	`.jsx`	FILE node only	`TreeSitterJSXParser`
Python	`.py`	regex	`TreeSitterPythonParser`
Go	`.go`	regex	`TreeSitterGoParser`
Rust	`.rs`	regex	`TreeSitterRustParser`
Java	`.java`	regex	`TreeSitterJavaParser`

Tree-sitter grammars are optionalDependencies. Missing grammars fall back silently per language.

Parsers extract: functions, arrow-function consts, methods (with scopePath), classes, interfaces, type aliases, enums, imports.

Parser source files

src/parsers/tree-sitter-typescript-parser.ts — TS/TSX/JS/JSX parsers
src/parsers/tree-sitter-parser.ts — Python/Go/Rust/Java parsers
src/parsers/typescript-parser.ts — regex fallback for TypeScript

MAGE Algorithms

The server uses Memgraph MAGE (memgraph/memgraph-mage:latest) for native graph algorithms:

Algorithm	MAGE call	Fallback
Community detection (Leiden)	`CALL community_detection.get()`	directory heuristic
PageRank PPR	`CALL pagerank.get()` + 3-hop Cypher	JS power iteration
BM25 text search	`CALL text_search.search()`	lexical scorer

Result objects include a mode field ("mage_leiden", "directory_heuristic", "mage_pagerank", "js_ppr") indicating which path ran.

Engines

Engine	File	Responsibility
`GraphOrchestrator`	`src/graph/orchestrator.ts`	File discovery, parse dispatch, Memgraph writes, incremental/full rebuild
`ArchitectureEngine`	`src/engines/architecture-engine.ts`	Layer rules, violation detection, layer suggestion
`TestEngine`	`src/engines/test-engine.ts`	Test file detection, affected test selection, categorization
`ProgressEngine`	`src/engines/progress-engine.ts`	Task/feature CRUD, blocking issue detection
`EpisodeEngine`	`src/engines/episode-engine.ts`	Episodic memory, decision query, reflect synthesis
`CoordinationEngine`	`src/engines/coordination-engine.ts`	Agent claims, release, staleness detection
`CommunityDetector`	`src/engines/community-detector.ts`	Leiden / directory-heuristic communities
`EmbeddingEngine`	`src/vector/embedding-engine.ts`	Symbol embedding generation and lookup
`HybridRetriever`	`src/graph/hybrid-retriever.ts`	RRF fusion of vector + BM25 + graph expansion
`PPR`	`src/graph/ppr.ts`	Personalized PageRank for relevance ranking

Tool Surface (39 tools)

Graph/querying (4): graph_set_workspace, graph_rebuild, graph_health, graph_query

Code intelligence (5): code_explain, find_pattern, semantic_slice, context_pack, diff_since

Architecture (2): arch_validate, arch_suggest

Semantic/similarity (4): semantic_search, find_similar_code, code_clusters, semantic_diff

Test intelligence (5): test_select, test_categorize, impact_analyze, test_run, suggest_tests

Progress tracking (4): progress_query, task_update, feature_status, blocking_issues

Episode memory (4): episode_add, episode_recall, decision_query, reflect

Agent coordination (4): agent_claim, agent_release, agent_status, coordination_overview

Utility (1): contract_validate

API Endpoints

# MCP Streamable HTTP
POST http://localhost:9000/
POST http://localhost:9000/mcp

# Health
GET  http://localhost:9000/health
GET  http://localhost:9000/info

Environment Variables

MEMGRAPH_HOST=localhost          # default: localhost
MEMGRAPH_PORT=7687               # default: 7687
MCP_PORT=9000                    # default: 9000
LXDIG_PROJECT_ID=my-repo    # optional default project namespace
LXDIG_USE_TREE_SITTER=true  # enable tree-sitter AST parsers

Build & Run

# Install (optionalDependencies include tree-sitter grammars)
npm install

# Build TypeScript
npm run build

# Start HTTP server (port 9000)
npm run start:http

# Development watch mode
npm run dev

# Run tests
npm test

Docker Stack

# docker-compose.yml uses:
memgraph/memgraph-mage:latest   # Memgraph with MAGE algorithms
qdrant/qdrant:latest            # Vector database

Response Profiles

All tools accept profile parameter:

Profile	Description
`compact` (default)	Low-token, answer-first; omits verbose metadata
`balanced`	Moderate context with key supporting data
`debug`	Full unshaped payload for investigation

Repository Map

src/
  server.ts                    MCP HTTP surface (39 tools)
  index.ts                     legacy stdio entry
  config.ts                    environment config
  tools/
    tool-handlers.ts           all 39 tool implementations
  graph/
    orchestrator.ts            file discovery + parse dispatch + Memgraph writes
    client.ts                  Memgraph Bolt client
    builder.ts                 Cypher node/edge builders (SCIP IDs)
    hybrid-retriever.ts        RRF fusion retrieval
    ppr.ts                     Personalized PageRank
    watcher.ts                 file watcher for incremental rebuilds
    cache.ts                   query result cache
  parsers/
    tree-sitter-typescript-parser.ts   TS/TSX/JS/JSX tree-sitter parsers
    tree-sitter-parser.ts              Python/Go/Rust/Java tree-sitter parsers
    typescript-parser.ts               regex fallback for TypeScript
  engines/
    architecture-engine.ts
    test-engine.ts
    progress-engine.ts
    episode-engine.ts
    coordination-engine.ts
    community-detector.ts
    migration-engine.ts        Schema migration helpers (internal, not exposed as tools)
  vector/
    embedding-engine.ts
    qdrant-client.ts
  response/
    budget.ts                  token budget enforcement
  cli/                         CLI utilities
docs/
  GRAPH_EXPERT_AGENT.md        agent runbook
  AGENT_CONTEXT_ENGINE_PLAN.md implementation plan + phase status

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

lxDIG — Dynamic Intelligence Graph Architecture

Overview

Server Implementation

Key Architectural Properties

Data Planes

Graph Schema

Parser Architecture

Parser registry

Parser source files

MAGE Algorithms

Engines

Tool Surface (39 tools)

API Endpoints

Environment Variables

Build & Run

Docker Stack

Response Profiles

Repository Map

FilesExpand file tree

ARCHITECTURE.md

Latest commit

History

ARCHITECTURE.md

File metadata and controls

lxDIG — Dynamic Intelligence Graph Architecture

Overview

Server Implementation

Key Architectural Properties

Data Planes

Graph Schema

Parser Architecture

Parser registry

Parser source files

MAGE Algorithms

Engines

Tool Surface (39 tools)

API Endpoints

Environment Variables

Build & Run

Docker Stack

Response Profiles

Repository Map