AiDex

AiDex is an open-source MCP server from CSC-Software that gives AI coding assistants persistent memory, code search and live telemetry. It indexes a project once with Tree-sitter into a local SQLite database, so assistants can query identifiers and file signatures instead of grepping and reading whole files. It works with MCP-compatible clients such as Claude Code, Claude Desktop, Cursor, Windsurf, Gemini CLI and VS Code Copilot.

What It Is

AiDex is a code-intelligence and memory layer for AI coding agents, distributed as the npm package aidex-mcp and run as a local MCP server. Running aidex_init on a project creates a .aidex/ folder containing an index.db SQLite database. The index covers identifiers, methods, classes and project files across 14 languages (C#, TypeScript, JavaScript, Rust, Python, C, C++, Java, Go, PHP, Ruby, HCL/Terraform, Kotlin, Swift), plus Astro frontmatter. The project states that it is 50x more token-efficient than grep, and the README shows an example of about 50 tokens versus 2,000+ for a lookup.

Search and Index Workflow

Identifier queries support exact, contains and starts_with modes, with time filters such as modified_since (for example 2h) to find recent changes. Signature tools return the classes and methods of a file or glob without reading the full file. Optional semantic search embeds code, Markdown docs and workspace items (tasks, notes) locally and offers exact, semantic and hybrid (RRF fusion) modes. An opt-in LLM layer (Anthropic, OpenAI, OpenRouter, Ollama or HuggingFace) can translate, expand and rerank queries. The llm_send_code privacy switch defaults to off, so only query text and metadata are sent. Global tools search across every indexed project.

Memory, Tasks and Guidelines

Session notes persist between chats and are archived with searchable history. A built-in task backlog supports priorities, tags, auto-logged history and scheduled or recurring tasks. Global guidelines store coding conventions shared across projects.

Log Hub, Dashboard and Screenshots

The Log Hub receives logs from any program over HTTP POST (no SDK), which the AI can query and which stream live to a browser Viewer. A Debug Dashboard shows fixed-slot widgets (label, progress, gauge, plot), and a Control API lets the AI send commands back to a running app. A screenshot tool captures screens or windows with scale and color reduction to cut token use. The Viewer at localhost also offers a file tree, signatures, search, tasks and settings tabs.

Setup Path

Install with npm install -g aidex-mcp on Node.js 20 or later. Post-install setup detects supported AI clients and registers AiDex as an MCP server, and manual configuration snippets are provided for each client.