Local, token-aware context engine for AI coding assistants
ndxr by Ndxr is a local-first context engine for the Model Context Protocol that supplies AI coding agents with tightly scoped code inputs. It produces token-budgeted context for developer tasks and reduces irrelevant material sent to models. The tool emphasizes fast indexing, intent-aware retrieval, and execution-path analysis, aimed at software engineers who use AI assistants and require private, efficient access to large, multi-language repositories and ongoing development sessions.
What tasks can you actually use ndxr for?
ndxr targets multi-file developer work where an agent needs persistent context across sessions. The engine provides logic flow tracing to find execution paths between symbols, impact analysis that maps a change's blast radius, and session memory that persists AI observations and decisions across separate sessions, which helps agents resume complex refactors or debugging without reprocessing entire repositories.
How reliable are the retrieved code contexts for token-limited models?
Instead of returning whole files, the index operates on symbols and edges such as calls, imports, and dependencies, so snippets focus on structural relevance. The search pipeline combines BM25 relevance and PageRank centrality with optional semantic embeddings, and Context Capsules pack related symbols into a user-defined token budget, which reduces token waste and keeps returned context within model limits.
Is it easy to integrate into an existing coding workflow?
Integration includes a command that sets up .mcp.json and CLAUDE.md for MCP clients, and ndxr ships as a single static binary for Linux, macOS, and Windows. A live file watcher updates the index in real time and incremental indexing updates changed files in less than a second, so the index remains current during active development without full reindexes.
Does it handle private code and local processing?
All parsing, indexing, and searching run on the local host and require no API keys or cloud services, so source code does not leave the machine. That execution model keeps control and auditability within the developer's environment, which suits teams that must avoid cloud transfers while using AI agents against large repositories.
A focused choice for MCP-centered developer teams
As an open-source project built for the MCP ecosystem and noted for high-speed performance powered by Rust and Tantivy, ndxr fits teams adopting MCP-capable assistants who prioritize local control and precise context delivery. Expect a tool oriented toward code-centric AI workflows rather than general-purpose code search. A practical tip: pair ndxr retrievals with human review during complex refactors to confirm semantic intent.
