Offline MCP server for AI coding agents with semantic code search, symbol intelligence, dependency graphs, and safe file operations.

Lumora

One MCP server. Full codebase intelligence. Complete file operations.

Lumora gives AI coding agents everything they need to understand and modify a codebase — semantic code search, dependency graphs, symbol lookup, file reading, editing, and creation — through a single MCP connection.

No cloud. No API keys. Just cargo install lumora and go.

The Problem

AI coding agents are powerful, but they're flying blind. They cat entire files, grep for strings, and hope for the best. The result: bloated context windows, missed connections, and wasted tokens.

Without Lumora, your agent:

• Reads entire files when it only needs 10 lines
• Can't tell you what calls a function without scanning every file
• Has no concept of dependency paths between modules
• Burns tokens on irrelevant code just to find what it needs

With Lumora, your agent:

• Reads exactly the lines it needs with automatic truncation
• Finds all callers, references, and definitions through a semantic graph
• Traces dependency paths between any two files or symbols
• Gets minimal, high-signal context slices — not entire files

Install

cargo install lumora

Requires a C toolchain for tree-sitter and bundled SQLite:

• Linux: sudo apt install build-essential (or equivalent)
• macOS: xcode-select --install
• Windows: Visual Studio C++ Build Tools

Quick Start

Lumora auto-detects your repository root from the current directory. No --repo flag needed.

# Index your codebase (run from anywhere inside the repo)
lumora index

# Search the graph
lumora query symbol main
lumora query refs index_repository --order score_desc --limit 50

# Start the MCP server
lumora mcp

MCP Setup

Lumora works with any MCP-compatible coding tool. Pick your client below.

Claude MCP add lumora lumora mcp

Add to `opencode