abhigyanpatwari/GitNexus
Developer ToolsIndex a codebase into a queryable knowledge graph that coding agents and editors can search via MCP.
GitNexus
Overview
GitNexus parses a repository into a graph of symbols, calls, dependencies and execution flows. The index is local and the CLI does not call an LLM, so the core analysis works without an API key. Coding agents reach the graph through MCP tools; developers can also use a CLI or a web graph explorer. It is aimed at teams that want their assistant to answer questions like "what calls this function?" or "what would this change affect?" against a persistent index instead of re-reading files.
The same index backs several operations: query for code and execution flows, context for a symbol's callers and dependencies, impact for change blast radius, trace for paths between symbols, detect_changes for reviewing a Git diff, and rename for a previewed semantic rename.
Quick start
Node.js 22.18+ on the 22.x line, or 24.11+, is required. Run the commands from the repository root you want to index:
npx gitnexus analyze
npx gitnexus setup
analyze builds the index and also installs agent skills, registers Claude Code hooks and writes AGENTS.md / CLAUDE.md. Use gitnexus analyze --index-only if you want the index without those repository files. setup configures MCP for supported editors; to pick specific ones, pass them, for example gitnexus setup -c cursor,codex. See the editor setup guide for the support matrix and manual configuration, since Windsurf needs manual MCP setup.
After indexing, common commands include gitnexus status to check index freshness, gitnexus query "authentication" to search the graph, and gitnexus impact UserService --direction upstream. For several clients, run one shared server with gitnexus mcp --http --port 3111 and point clients at http://127.0.0.1:3111/mcp rather than using stdio per client.
To browse visually, run gitnexus serve and open the web UI; it uses the existing index without re-uploading or re-indexing. Full command and flag details are in the CLI reference.
Supported languages
Registered providers cover C, C++, C#, COBOL, Dart, Go, Java, JavaScript, Kotlin, Objective-C, PHP, Python, Ruby, Rust, Swift, TypeScript, Vue and Zig. Coverage varies by construct, and the language inventories document measured behavior, exclusions and gaps. Resolution quality also depends on available evidence, so the agent should be expected to report ambiguity and index freshness limits rather than give certainty.
Constraints and deployment
- The OSS version is under the PolyForm Noncommercial 1.0.0 license; commercial use requires a separate license, and an enterprise SaaS or self-hosted offering covers PR review, multi-repo graphs and OCaml support.
- The CLI indexer runs locally without an LLM. Optional remote embeddings, wiki generation and AI chat use whichever providers you configure, so data leaves your machine only through those services.
- If you expose an index to other users through the web UI or a hosted deployment, read the security guidance and the deployment authentication notes first; Docker and hosted deployment steps live in the deployment guide.
- Native dependencies on Linux and Windows are listed separately in the platform requirements, and installation issues are covered in troubleshooting.
HotRepo's take
GitNexus appears in HotRepo's daily all-language snapshots for 28–31 August 2026, when it moved from 46,074 to 46,714 stars with daily gains between 181 and 273. No all-language daily rank has been recorded for it, and the visible records are a short run rather than a long series, so they support only the observation that the project was gaining steadily over those four days. The more concrete signal is the product shape: an offline, locally indexed graph with MCP integration and a browser explorer, plus a paid enterprise tier for PR blast-radius review and multi-repo graphs. That combination targets teams already using coding agents who want repository-wide context; small or single-language projects may get less value than the setup cost.
Written by AI from the project's README, updated . The project's own docs are the source of truth.