GitNexus

A browser based tool that indexes a repository into a knowledge graph of dependencies, call chains, and execution flows, then exposes it through MCP tools for AI agents.

The GitHub description calls GitNexus a zero server code intelligence engine that runs entirely in the browser. The README says it indexes any codebase into a knowledge graph covering dependencies, call chains, clusters, and execution flows, then exposes it through MCP tools so AI agents never miss code. The project lists TypeScript as its primary language. Keeping everything client side is the headline design decision, since it removes the need to run any infrastructure to get the graph.

The README opens with an unusual notice: GitNexus has no official cryptocurrency, token, or coin. It warns that any token using the name on pump platforms is not affiliated with, endorsed by, or created by the project or its maintainers, and advises against purchasing such cryptocurrency. Project owners rarely have to write that paragraph, and its presence here says something about what has been happening around projects like this one.

Two commands cover the setup. An analyze command indexes the codebase, installs agent skills, registers hooks, and creates context files in one step, while a setup command writes the MCP config. The README documents an npm 11.x issue with npx and recommends pnpm, and it notes that installing globally writes an absolute path MCP config that bypasses npx. That level of install detail is the kind of thing a maintainer writes after running into the problem themselves.

There is a CLI plus MCP mode, recommended for daily development with Cursor, Claude Code, and Codex, and a Web UI for quick exploration and demos. The CLI indexes full repositories of any size, while the web UI is limited by browser memory unless backend mode is used. A bridge mode connects the two. The split lets someone use the lightweight web view for demos and the CLI for real work.

The README argues that coding agents do not truly know a codebase structure, giving an example where a change breaks dependencies. It says traditional graph RAG gives raw edges while GitNexus precomputes structure at index time, providing reliability, token efficiency, and model democratization so smaller LLMs work. The claim is that doing the structural work once at index time is cheaper and more reliable than having an agent rediscover it on every query.

GitNexus bets that precomputed structure beats on demand exploration, and runs in the browser to lower the barrier to trying it. The cryptocurrency warning is an unfortunate sign of the times, but the project stays on the code intelligence side.

• Source 1 · Project repository
• Source 2 · README