Frequently Asked
Questions

Devscriptor is a high-performance code analysis engine built in Rust that builds a knowledge graph of your codebase in a single SQLite file. It extracts entities (functions, classes, etc.), relationships (calls, inheritance), enables semantic search, and provides LSP code intelligence—making your codebase queryable by AI agents through the Model Context Protocol (MCP).

For individual and non-commercial use, yes — permanently. Personal projects, learning, side projects, and open-source work cost nothing, with no license key, no account, and no expiry. Commercial use requires a paid seat at $19 per developer per month (or $190 per year). Both tools are distributed as pre-built installers for Linux (x86_64 and aarch64), macOS (Apple Silicon and Intel), and Windows (x86_64) from downloads.devscriptor.com. Just download, extract, and run.

Yes, with a seat. Commercial use — company code, paid client work, or a commercial product or service — costs $19 per developer per month, or $190 per developer per year. A seat is counted per person, not per machine: laptops, dev containers, CI runners, and build agents are unlimited. Evaluating the tools inside a company, before you commit, is free. Individual and non-commercial use stays free permanently. See the Commercial License page to buy seats or read the detail.

No. There is nothing to activate, nothing to deploy, and no phone-home. The binaries are byte-identical whether or not seats have been bought — the commercial license is a legal agreement, not a technical lock. This is deliberate: enforcing a license means adding a network callback, and we are not going to ship one to paying customers when the free build has none. It also means no trial timer, no offline grace period to manage, and nothing that breaks when a CI runner has no outbound internet.

Devscriptor includes a native, unified MCP (Model Context Protocol) server that exposes 75 tools over JSON-RPC on stdio to AI agents like Claude Desktop, Cursor, and Cline. This includes 32 core tools for scanning, search, analysis, the Context Graph, and the real-time watcher; 38 LSP tools for live code intelligence (go-to-definition, hover, completions, diagnostics); and 5 architecture tools for LSP detection and setup. Configure your client with command "devscriptor" and args ["mcp"] — one binary, one name, on every platform.

Yes. Devscriptor is an MCP server, not an LLM — it works with any MCP-compatible client, including agents backed by locally hosted models. Because all of Devscriptor’s own analysis runs on your machine, nothing about your code depends on a cloud provider. Your AI client (Claude Desktop, Cursor, Cline, etc.) connects to Devscriptor over stdio and can route to whatever model you choose.

No. Your code never leaves your machine. Scanning, parsing, indexing, and search make no network calls, all data lives in a local SQLite file, and there is no telemetry, no analytics, and no tracking of any kind. Devscriptor only touches the network to fetch things you ask for: the one-time embedding-model download on your first embed, and on-demand LSP server installs. Neither transmits your code, and both can be pre-staged so the tool runs fully air-gapped. This makes it well suited for proprietary codebases and environments with strict data-security requirements.

Devscriptor parses 26 languages and frameworks with tree-sitter 0.25: Python, Rust, TypeScript/TSX, JavaScript, Go, Java, C, C++, C#, Ruby, PHP, Swift, Kotlin, Scala, Dart, Bash, SQL, Solidity, F#, GraphQL, HTML, Objective-C, CUDA, and VB.NET — plus the Vue and Svelte frameworks. It extracts functions, classes, structs, enums, interfaces, components, and their relationships (calls, inherits, implements, uses).

LSP (Language Server Protocol) integration provides real-time IDE features: go-to-definition, hover information, completions, diagnostics, call hierarchy, and type hierarchy. Devscriptor includes 38 LSP tools and 5 architecture tools, and ships 21 pre-configured language servers resolved via a 4-tier detection system (user config → Devscriptor-managed → system PATH/SDKs → VS Code extensions), then installed on demand. Servers include rust-analyzer, pyright, typescript-language-server, gopls, clangd, and more.

Context Graph is an AI memory system that persists across sessions in the same local SQLite database as the code graph. It stores notes using 9 structured types: adr (architecture decisions), standard (coding standards), optimization, failure, practice, reflection, general, file_description, and modification. Notes can be linked to concrete code entities and abstract concepts, and retrieved by keyword, semantic, or hybrid search. When you return to a project weeks later, your AI recalls the context.

Devscriptor generates 768-dimensional vector embeddings for your code using the local jina-embeddings-v2-base-code model (INT8-quantized, ~150 MB). No API keys are required — the model downloads once, then runs offline. These embeddings capture the semantic meaning of code, so you can search by concept (e.g., "functions that handle authentication") rather than keywords. Vector search uses cosine similarity via sqlite-vec, and can be fused with FTS5 keyword results using Reciprocal Rank Fusion.

Yes. Devscriptor is built in Rust and uses parallel, gitignore-aware file discovery, memory-mapped parsing, and a streamed cross-file analysis pass with bounded memory. Incremental scanning (mtime+size+XXH3) and a content-hash embedding cache make re-analysis fast. The streaming cross-file pipeline has been validated on large repositories such as Kubernetes (~11.5k Go files) without a memory cliff.

Download the installer for your platform from downloads.devscriptor.com, extract it, and add the devscriptor binary to your PATH. Then run "devscriptor scan ./your-project" to build the knowledge graph, and optionally "devscriptor embed" for semantic search. For AI integration, configure your MCP client with command "devscriptor" and args ["mcp"]. See the Download page for full setup instructions.

Rust provides memory safety without garbage collection, zero-cost abstractions, and predictable performance. Devscriptor leans on that with multi-threaded parsing (scaling to num_cpus), memory-mapped file reading, a pure-Rust ONNX embedding engine (tract) with a RAM- and core-aware auto-tuner, batched WAL-mode SQLite transactions, and content-hash caches that skip redundant work. The result is a fast, low-overhead engine that stays responsive on large codebases while keeping strong safety guarantees.