Rig: Local code graph for coding agents, with a single npx command

A repository published this week on Hacker News is called Rig and proposes something concrete: build a graph of a project's source code locally, without relying on any external service, and make it available to coding agents with a single `npx rig` command in the terminal. The proposal appeared in the HN community on May 28, 2026 with few points and no comments yet, indicating it is very recent and has not had time to accumulate public discussion.

The premise is straightforward: coding agents, whether Claude Code instances, custom sub-agents, or any other system that consumes code context, work better when they understand the structure of the project, not just the literal content of the files. A code graph captures relationships between functions, modules, and imports, allowing the agent to reason about dependencies without having to read thousands of raw lines.

What Rig does exactly

According to the repository description, Rig analyzes the source code in the current directory and builds a graph of dependencies and symbols that remains available locally. Startup requires no global installation: just run `npx rig` in the project directory. The local-first approach means that no data leaves the machine, something that in environments with proprietary or sensitive code is far from a minor detail.

The project is presented as a piece designed to integrate with coding agents, though public documentation at the time of writing is still sparse. It's not entirely clear whether it exposes an API compatible with MCP (Model Context Protocol), generates a static context file, or spins up a local server the agent can query. These are open questions the community will start asking as soon as the project gains traction.

Why the local-first approach matters in 2026

In the current ecosystem, coding agents consume context from multiple sources: MCP servers, skills, project files, conversation history. The bottleneck is rarely the model's capacity—Claude Opus 4.7 handles up to one million tokens of context window—but rather the quality and relevance of the context delivered to it. Passing a structured code graph instead of flat files can reduce noise and allow the agent to reach the code it needs faster.

Rig's key differentiator over similar solutions, some of which already do something comparable through code indexing MCP servers, is frictionless installation. An `npx` removes the configuration barrier that causes many developers to abandon a tool before trying it. If the graph can be generated in seconds and consumed from Claude Code or any agent that reads context files, the use case becomes immediately accessible.

Who should consider trying it now

In its current state, Rig seems better suited for individual developers or small teams wanting to experiment with agents on medium-sized codebases. There is no evidence yet that it supports large monorepos or has been tested in complex production projects. Language support is also unclear: code graphs are more or less trivial to build in JavaScript/TypeScript, considerably more complex in languages with richer type systems or intensive metaprogramming.

What it does have going for it is timing: integrating external tools with Claude Code via MCP and plugins is in full swing, and any utility that reduces friction for delivering structured context to an agent has a place in many teams' workflows.

---

Our take: It's too early to know whether Rig fills a real gap that other code indexing MCP servers don't already cover, but the zero-installation and strictly local processing approach is a reasonable bet. It's worth following once it accumulates some real-world use and more complete documentation.