CodeGraph is a pre-indexed code knowledge graph that runs locally and wires into AI coding agents — Claude Code, Cursor, Codex, Gemini, OpenCode, AntiGravity, Kiro, GitHub Copilot, and Hermes Agent — so those agents query a semantic map of the codebase instead of reading files one at a time.

What it is

CodeGraph builds and maintains a graph of a codebase ahead of the moment an agent needs it. The index updates automatically when code changes, and agents reach it through MCP tools and a bundled CLI, so a request can resolve to the relevant definitions, call sites, and routes rather than walking the tree. According to the project README, the kernel is powered by Rust, and the distribution is self-contained: the release bundles its own runtime, so no Node.js installation and no native build step are required.

The concrete problem it solves is context cost. Agent workflows typically spend tokens and tool calls probing a repository file by file until they happen to land on the relevant code, and the project frames its benefit as fewer tokens and fewer tool calls for the same task. CodeGraph replaces that exploratory read-and-grep loop with a pre-built index that answers structural questions directly. The README states the tool is 100% local, so source does not leave the machine as part of building or querying the graph.

Key capabilities

• Pre-indexed code knowledge graph that auto-syncs as code changes, so the graph stays current without a manual rebuild.
• MCP tools exposed to supported agents, documented under an MCP Tools section in the README.
• CLI lifecycle commands including codegraph upgrade to update in place, codegraph upgrade --check to test whether an update exists, and a pinned-version upgrade form.
• The upgrade path detects whether the install came from the bundle, npm, or npx, and updates the matching installation.
• Framework-aware route detection, so framework routes surface as graph nodes rather than plain string literals.
• Mixed iOS / React Native / Expo bridging, covering projects that span native and JavaScript layers.
• Three install routes: curl -fsSL https://raw.githubusercontent.com/colbymchenry/codegraph/main/install.sh | sh on macOS and Linux, irm https://raw.githubusercontent.com/colbymchenry/codegraph/main/install.ps1 | iex on Windows, or npm i -g @colbymchenry/codegraph where Node is present.

Who uses it and how

• Developers working across more than one agent: the supported list spans nine tools, and each can be wired to the same local graph.
• Mobile teams whose projects bridge native iOS and React Native or Expo code, which the README addresses as a named scenario.
• Teams on framework-driven web applications that want route structure available to the agent rather than inferred from searches.
• Cross-platform development groups, since Windows, macOS, and Linux are all listed as supported platforms.
• Projects large enough that cross-file coverage is measured and reported in the README, where per-file reads strain the context budget.

Getting started

Install with the shell command for macOS and Linux, the PowerShell command for Windows, or npm i -g @colbymchenry/codegraph if Node is already present. The installer places codegraph on the PATH without modifying the current shell, so a new terminal is needed before wiring up an agent.

How it compares

The facts provided name no paid products that CodeGraph replaces and no similar tools in this category, so no direct contrast is possible here. On the evidence available it stands alone in this registry, and its only documented comparison point is the unassisted agent workflow it is designed to make cheaper.

When to use it — and when not to

Pick CodeGraph when agents are already part of the workflow, when the codebase is large enough that repeated file reads dominate token spend, and when keeping the index on local machines is a requirement rather than a preference. The trade-offs are visible in the metadata: the repository carries 518 open issues, and the hosted platform described in the README is still at the early beta access stage, so anyone wanting managed pull-request analysis cannot get it today. Skip it if the agent in use is not on the supported list, or if the project needs a hosted service now rather than a local index each developer must run. The MIT licence is unambiguous, which removes licence risk from the decision.

# macOS / Linux
curl -fsSL https://raw.githubusercontent.com/colbymchenry/codegraph/main/install.sh | sh

# Windows (PowerShell)
irm https://raw.githubusercontent.com/colbymchenry/codegraph/main/install.ps1 | iex

npm i -g @colbymchenry/codegraph

codegraph install

cd your-project
codegraph init

codegraph uninstall