codeindex

codeindex builds a persistent, queryable knowledge graph of a codebase rather than re-scanning files on every query: pointed at a project (Python, JavaScript/TypeScript, Go, Ruby, Rust, Java, PHP, and more), it produces a SQLite graph store, a dependency index, and a symbol map, all incremental and temporal — meaning you can query not just the current state but historical “as-of” states derived from git history.

Its headline feature is blast-radius scoring: for any file, it estimates how many other files would break if that file changed, backed by both current and historical analysis. Hybrid semantic search lets you find symbols using natural language fused with keyword and graph-based expansion, rather than exact-match grep alone.

Apache-2.0 licensed and distributed via PyPI (pip install codeindex), the tool exposes its data ten different ways — CLI, markdown reports, an MCP server with 10 tools, a pre-commit hook, CLAUDE.md injection, and an interactive 2D/3D visualization UI — with zero required runtime dependencies since SQLite is part of Python’s standard library.

What You Get

• A persistent, incremental SQLite graph store of your codebase’s dependencies and symbols
• Blast-radius scoring showing how many files break if a given file changes, including historical as-of queries
• Hybrid semantic search over symbols, combining natural language with keyword and graph expansion
• Ten ways to consume the index: CLI, markdown report, MCP server (10 tools), pre-commit hook, CLAUDE.md injection, and a visualization UI

Common Use Cases

• Assessing the blast radius of a proposed change before making it, across current and historical codebase states
• Giving an AI coding assistant a symbol index so it can find relevant code without scanning every file
• Adding a pre-commit hook that flags high-impact changes based on dependency graph analysis
• Visualizing a large codebase’s dependency structure through 2D/3D graphs, a dependency matrix, or a treemap

Under The Hood

Architecture codeindex persists its analysis to a SQLite database at <repo>/.codeindex/index.db rather than recomputing everything per invocation, with a separate codeindex.json dependency index and symbolindex.json symbol map written into the repo for interoperability. The temporal aspect comes from backfilling the graph from git commit history without touching the working tree, letting blast-radius and dependency queries be answered “as of” any historical point, not just the current HEAD.

Tech Stack Python, using SQLite (Python’s standard library) as the sole persistence layer to keep runtime dependencies at zero, with multi-language static analysis support (Python, JS/TS, Go, Ruby, Rust, Java, PHP). An MCP server exposes 10 distinct tools for agent consumption, and an interactive visualization UI renders the graph as 2D/3D views, a dependency matrix, or a treemap.

Code Quality Very active, consistently maintained commit history and a genuinely broad output surface (10 distinct consumption modes) suggest deliberate design for integration into varied workflows rather than a single-purpose CLI; the zero-required-dependency design choice (SQLite via stdlib) reflects attention to installation friction.

What Makes It Unique Most code-dependency tools give you a snapshot of current-state relationships; codeindex’s temporal, git-history-backed graph specifically supports historical blast-radius queries — understanding not just what a change affects today, but how that impact would have looked at any past commit — which is unusual among AI-assisted-development tooling.

Self-Hosting

Licensing Model Apache-2.0 licensed — fully open source with no license key.

Self-Hosting Restrictions Not applicable; it’s a local CLI/library tool with a local SQLite store, no hosted service.

License Key Required No.