🧠 KacheDB MCP Server

High-Performance Model Context Protocol (MCP) server for KacheDB — exposing sub-millisecond in-memory caching, Tree-Sitter AST symbol intelligence, and SIMD semantic vector memory to Antigravity IDE, Claude Desktop, Cursor, and AI coding agents.

⚡ Why KacheDB for AI Agents?

AI coding assistants repeatedly re-read large codebases, parse ASTs, and crawl directories, wasting tens of thousands of tokens and adding hundreds of milliseconds of latency per turn.

kachedb-mcp connects your AI assistant directly to KacheDB's Megaslab pure-RAM SwissTable:

• 🚀 Sub-50 Nanosecond Retrieval: Instant symbol inspection and exact KV hits in memory.
• 🌳 Tree-Sitter AST Code Intelligence: Multi-language symbol graphs (functions, classes, structs, signatures, docstrings) across Python, Rust, Go, TypeScript, and JavaScript.
• 🔍 Upward Anchor Discovery: Zero-config project root detection from any touched file — no arbitrary $HOME scans or guessing folder names.
• 🔄 Incremental Git Watermarks: Sub-millisecond sync verification using commit SHA watermarks (kache:meta:<ws>:commit). Up-to-date projects are skipped with zero overhead.
• 🧠 SIMD Semantic Vector Memory: Natural language concept and code recall powered by ARM NEON & AVX2/FMA cosine similarity kernels.
• 🪙 Live Financial Telemetry: Tracks avoided latency, cumulative tokens saved, and real-time USD cost savings.

🛠️ MCP Tools Exposed (11 Tools)

🌳 Codebase Intelligence Tools

🧠 Semantic Memory Tools

⚡ Exact Key-Value Tools

📊 Telemetry & Health Tools

🚀 Quickstart

1. Start the KacheDB Server Daemon

Ensure your KacheDB daemon is running locally:

kachedb-server --port 6380

2. Configure in Antigravity IDE / Claude Desktop / Cursor

Add kachedb to your MCP configuration file (mcp_config.json or claude_desktop_config.json):

Zero-Config Mode (Recommended)

No hardcoded paths needed. When an agent opens or mentions any file, Upward Anchor Discovery automatically resolves the enclosing project root and indexes it on-demand:

{
  "mcpServers": {
    "kachedb": {
      "command": "uvx",
      "args": ["kachedb-mcp"],
      "env": {
        "KACHEDB_PORT": "6380"
      }
    }
  }
}

Pre-Configured Multi-Project Roots

To automatically discover and index projects on startup, set KACHEDB_WORKSPACE_ROOTS:

{
  "mcpServers": {
    "kachedb": {
      "command": "uvx",
      "args": ["kachedb-mcp"],
      "env": {
        "KACHEDB_PORT": "6380",
        "KACHEDB_WORKSPACE_ROOTS": "/path/to/projects:/path/to/work"
      }
    }
  }
}

Or install via Python pip:

pip install kachedb-mcp

{
  "mcpServers": {
    "kachedb": {
      "command": "kachedb-mcp",
      "env": {
        "KACHEDB_PORT": "6380"
      }
    }
  }
}

🌐 Multi-Language AST Support

Tree-Sitter parsers extract rich symbol graphs with zero external compiler dependencies:

⚙️ Environment Configuration

📄 License

Licensed under either of Apache License, Version 2.0 or MIT License at your option.