Free to install and use

Download Devscriptor

One universal binary for your platform — CLI, LSP integration, Context Graph, and the unified MCP server, all in a single download. No package managers, no build-from-source.

Linux x86_64 / aarch64 macOS (Apple Silicon) Windows x86_64

Install from the archive

Every archive contains the devscriptor binary plus a devscriptor-mcp entry point (a symlink on Linux/macOS, a .cmd shim on Windows).

macOS & Linux

# Example: Linux x86_64 (swap the target for your platform)
curl -LO https://downloads.devscriptor.com/installers/devscriptor-1.2.0-x86_64-unknown-linux-gnu.tar.gz
tar -xzf devscriptor-1.2.0-x86_64-unknown-linux-gnu.tar.gz
sudo mv devscriptor-*/devscriptor devscriptor-*/devscriptor-mcp /usr/local/bin/
devscriptor --version

On macOS, if Gatekeeper blocks the first launch, clear the quarantine flag: xattr -d com.apple.quarantine /usr/local/bin/devscriptor

Windows (PowerShell)

# Download and extract the .zip, then add the folder to your PATH.
# The archive contains devscriptor.exe and a devscriptor-mcp.cmd entry point.
Expand-Archive -Path devscriptor-1.2.0-x86_64-pc-windows-msvc.zip -DestinationPath C:\devscriptor
devscriptor --version

First-time setup

Devscriptor separates a fast structural scan from optional semantic embedding, so the scan runs in seconds on any hardware.

Scan your codebase

Fast structural scan — FTS5 keyword search and the entity/relationship graph are ready in seconds. No model loaded.

devscriptor scan ./my-project

Add semantic search (optional)

Generate vector embeddings once — the model downloads on first run (~150 MB), then unloads. Requires ≥ 4 GB free RAM.

devscriptor embed

Set up language servers (optional)

Detect your project’s languages and install the LSP servers it needs via the 4-tier resolution system.

devscriptor lsp setup

Connect your AI agent

Point Claude Desktop, Cursor, or Cline at the devscriptor-mcp entry point to expose all 74 MCP tools.

devscriptor-mcp

System requirements

RAM (scan)No requirement — structural scan never loads the model

RAM (embed)4 GB free minimum, 8 GB recommended

CPU2 cores minimum, 4+ recommended

Disk~1 GB for the embedding model cache (2 GB recommended)

LinuxAny mainstream glibc distribution — no OpenSSL or extra packages

macOSApple Silicon (aarch64) only — Intel Macs are not supported

WindowsWindows 10 / 11, 64-bit

100% local & private. Scanning, indexing, and search make no network calls. The only outbound traffic is the one-time embedding-model download (Hugging Face, with the Devscriptor mirror as a fallback) and on-demand LSP server installs — both of which can be pre-staged for fully air-gapped use.