mcp-sqlite-graph-memory

Universal Model Context Protocol (MCP) server delivering dynamic SQLite-backed code knowledge graph memory for AI coding assistants and autonomous agents.

Automatically detects active workspaces, parses multi-language ASTs and structural dependencies into an isolated local SQLite knowledge graph, and exposes precision MCP tools to inspect symbols, calculate shortest call/import paths, and navigate code topologies.

🌟 Key Capabilities

• Zero Runtime Dependencies: Written entirely in Python standard library (ast, sqlite3, json, re, hashlib, threading).

• Dynamic Multi-Workspace Isolation: Generates deterministic SHA-256 hashes per workspace root; stores isolated SQLite databases in ~/.mcp-graph-memory/storage/graph_<hash>.db.

• Incremental Stat Caching: Verifies file mtime and size to achieve sub-millisecond skips for unchanged files during re-indexing.

• Multi-Language AST & Structural Parser:

  • Python: Classes, methods, functions, async functions, decorators, imports (from x import y, import z), inheritance hierarchies, call sites.

  • TypeScript / JavaScript: Classes, interfaces, types, functions, arrow functions, ES6 imports/exports, CommonJS require().

  • Go: Packages, imports, structs, interfaces, functions, methods.

  • Rust: Modules, structs, enums, traits, impl blocks, fn signatures.

  • C / C++ / Java / PHP / Ruby / Kotlin / Swift / Shell: Structs, classes, definitions.

• Freeze & CPU Starvation Protection:

  • Non-blocking asynchronous daemon initialization.

  • Non-busy stdio transport (for raw_line in sys.stdin:).

  • Strict ignore list for node_modules/, .git/, .venv/, binary weights, and cache blobs.

Related MCP server: Synapse

🛠️ MCP Tools Specification

⚡ Token Minimization & Compact Mode

All query tools support a format parameter:

• format: "compact" (Recommended for AI agents): Formats symbols and graph edges into SCIP-style single-line text (e.g. [class] OrderService @ service.py#L16-L21 | sig: class OrderService), achieving up to 75% reduction in LLM context token consumption.

• format: "json": Serializes responses into compact, unindented single-line JSON, cutting whitespace overhead by ~40%.

🚀 Installation

git clone https://github.com/Luiguard/mcp-sqlite-graph-memory.git ~/.mcp-graph-memory
chmod +x ~/.mcp-graph-memory/server.py

🔌 Integration Guides

1. Google Antigravity Integration

Option A: Global Configuration (Recommended)

Add graph-memory to your global Antigravity MCP configuration file at ~/.gemini/config/mcp_config.json:

{
  "mcpServers": {
    "graph-memory": {
      "command": "python3",
      "args": [
        "/home/YOUR_USER/.mcp-graph-memory/server.py"
      ]
    }
  }
}

(Replace /home/YOUR_USER with your actual $HOME directory path or use an absolute path).

Option B: Workspace-Level Configuration

For per-project activation, place mcp_config.json inside your workspace root under .agents/mcp_config.json:

{
  "mcpServers": {
    "graph-memory": {
      "command": "python3",
      "args": [
        "${workspaceFolder}/.mcp-graph-memory/server.py"
      ]
    }
  }
}

2. Claude Desktop Integration

Locate your Claude configuration:

• Linux: ~/.config/Claude/claude_desktop_config.json

• macOS: ~/Library/Application Support/Claude/claude_desktop_config.json

• Windows: %APPDATA%\Claude\claude_desktop_config.json

Add the server definition:

{
  "mcpServers": {
    "graph-memory": {
      "command": "python3",
      "args": [
        "/path/to/mcp-sqlite-graph-memory/server.py"
      ]
    }
  }
}

3. Cursor / VSCode / Windsurf Integration

In .cursor/mcp.json or .vscode/mcp.json:

{
  "mcpServers": {
    "graph-memory": {
      "command": "python3",
      "args": [
        "/path/to/mcp-sqlite-graph-memory/server.py"
      ]
    }
  }
}

🔧 Environment Variables

🧪 Verification & Testing

Execute the automated integration suite:

python3 test_server.py

📄 License

MIT License - Copyright (c) 2026 Benjamin Leimer. See LICENSE for details.