Cartographer 🤖 Agent Open source

What it does

Cartographer maps and documents a codebase so that coding agents can navigate it without reading every file. The classic mode is a Claude Code plugin that orchestrates parallel Sonnet subagents: it scans the file tree with token counts, plans how to split the work across budgets, spawns subagents that each analyse a slice, then synthesises their reports into a docs/CODEBASE_MAP.md architecture map (file purposes, dependencies, data flows, navigation guides) and points CLAUDE.md at it. Re-running it checks git history and only re-analyses changed modules. A newer v2 adds a standalone CLI that indexes a repo into a local SQLite graph and compiles bounded context "briefs".

What is inside

The plugin ships the mapping skill and its orchestration prompts. The v2 CLI adds commands to index the graph, verify artifacts, view a summary, emit briefs around a path or symbol, run removal and completeness audits, manage evidence-backed notes, diff two graphs, and score whether agents used graph context before editing. A thin newline-delimited MCP wrapper exposes the index, brief, context, verify, audit, and diff operations as tools.

Works with

Claude Code, installed through its plugin marketplace. The v2 CLI and MCP wrapper are agent-agnostic but the documented plugin flow is Claude Code.

How to install

/plugin marketplace add kingbootoshi/cartographer
/plugin install cartographer

Maintenance and safety

The README states an MIT licence; the repository has no separate licence file. Last pushed in May 2026. The plugin spawns Sonnet subagents and can consume significant tokens on large codebases; the author suggests Haiku subagents for a cheaper, less accurate run. Token counting needs tiktoken installed via pip.

Who should use it

Developers who want an up-to-date architecture map generated from their own code, and agent builders who want a queryable code graph to feed bounded context into other tools.