AxiomCode Graph

AxiomCode Graph is a code-graph engine from AxiomCode Inc. that parses a repository and builds a knowledge graph of who calls what. AI coding agents and engineers query that graph through a CLI, an MCP server, and agent plugins instead of grepping. The engine is distributed on npm as @axiomcode/code-graph, and the source is published under the Functional Source License 1.1 with an Apache 2.0 future license.

What It Is

AxiomCode Graph resolves each call from the receiver's declared and inferred type, the way a compiler does, rather than matching names from syntax. The resulting graph is stored per language in SQLite, with tiers such as known edge, multi-inferred, library boundary, and ambiguous unknown. Unresolved calls stay explicit instead of being presented as established relationships. Java, TypeScript and Python are listed as stable, while JavaScript, C#, HTML and CSS are listed as beta.

Questions it answers

The CLI and MCP tools answer questions about a change, each as numbered places with source code and a verification line that re-checks every printed edge.

• impact shows callers, what a change reaches, and exercising tests.
• path shows the call chain from one symbol to another.
• tests selects the tests that uncommitted edits can reach.
• context explains how a task or feature works.
• link records where an unresolved call lands.
• graph renders the codebase as an interactive page.

Agent setup

Install the engine with npm (Node 22.13 or later and Python 3 required), then add the plugin for your agent. Plugins are documented for Claude Code, Codex CLI, Copilot CLI, Gemini CLI, Cursor, Windsurf and Devin CLI, and any MCP client can use a single config entry. Hooks that keep the graph current run in Claude Code, Codex, Cursor and Gemini CLI. The graph rebuilds in the background after edits.

Benchmarks (vendor-reported)

AxiomCode reports that on 748 held-out Defects4J bugs its selected tests included every bug-revealing test for 94.9% of bugs. It also reports higher call-resolution accuracy than other tree-sitter-based graph tools on Java and TypeScript, and up to 50% fewer tool calls when exploring a new codebase. These figures come from the vendor's own benchmarks.