{ "markdown": "

\n \n \n \n \"dekko\"\n \n

\n\n

\n \"CI\"\n \"PyPI\"\n \"Downloads\"\n \"Python\"\n \"License:\n \"Ruff\"\n

\n\n\n\n**dekko** is a fast, offline, dependency-free **static code map\ngenerator** and **codebase indexer for LLM coding agents**. It scans a\nrepository with [tree-sitter](https://tree-sitter.github.io/) (no model\ntokens spent parsing) and writes:\n\n- **`MAP.md`** — a human-readable map: a per-directory overview, an\n embedded architecture diagram, load-bearing/orchestrator rankings,\n then every file's functions/methods with signatures, doc lines, and\n who calls and is called by whom.\n- **`map.json`** — the same graph in machine-readable form.\n\nOn top of the map, dekko gives an agent a token-cheap way to answer\nquestions like \"what does this file contain,\" \"who calls this function,\"\nand \"what do I need to safely change this\" — without reading whole files.\nIt ships as a **CLI**, a **Claude Code `/map` plugin + MCP server**\n([Model Context Protocol](https://modelcontextprotocol.io/)), and\nworks with **Cline** too.\n\n## Why dekko?\n\nMost agent workflows gather context by reading whole files or grepping\nacross a repo — expensive, and it throws away structure (who calls\nwhat, what a function's fan-in/fan-out looks like). dekko instead\nparses the repo once into a call graph and answers targeted questions\nagainst it. Measured across 7 real, unmodified open-source repos\n(Go, TypeScript, Java, Rust, Python/C++ — up to 14k files), dekko's\nstructured queries used **3x–200x fewer tokens** than the equivalent\n`Read`/`Grep` workflow for the same task (repo orientation, outlining a\nlarge file, tracing a symbol's callers/callees).\n\n| Task | Example repo (scale) | dekko | Read/Grep | Savings |\n|---|---|---:|---:|---:|\n| Repo orientation (`summary`) | awesome-go (10 files) | 308 tok | ~15,271 tok | ~50x |\n| Repo orientation (`summary`) | cline (2,730 files) | 1,202 tok | ~4,020 tok | ~3.3x |\n| Outline a large file | claude-code `main.tsx` (4,683 lines) | 1,017 tok | 200,981 tok | ~197x |\n| Outline a large file | zed `editor.rs` (12,554 lines) | 1,996 tok | 115,109 tok | ~58x |\n| Symbol lookup (`query_symbol` + callers/callees) | tensorflow `Graph` class | ~811 tok | 61,656 tok | ~76x |\n| Symbol lookup (`query_symbol` + callers/callees) | spring-boot `prepareContext` | 759 tok | ~18,460 tok | ~24x |\n| Bundled context (`workset`) | zed | 2,984 tok | ~5,903+ tok (targeted) / ~164,571 tok (whole file) | ~2x / ~55x |\n| Bundled context (`workset`) | awesome-go | 617 tok | ~6,136 tok | ~10x |\n\ndekko's cost stays roughly flat per query while `Read`/`Grep` scales\nwith file/repo size, so the ratio grows with scale. The win isn't\nuniversal — small,\nself-contained files and already-grep-friendly local symbols see\nlittle to no benefit, and a few cases in the raw data are void because\nthe cheap answer was also an incomplete one. See\n[`benchmarks/real-world-repos/`](benchmarks/real-world-repos/README.md)\nfor the full per-task breakdown, methodology, and correctness caveats.\n\nCompared to tag-index tools like `ctags`/`gtags`, dekko resolves actual\ncall edges (not just definitions), ranks files by load-bearing-ness,\nand speaks directly to agents over MCP or the CLI — no editor plugin\nrequired.\n\n## Install\n\n```sh\nuv tool install dekko # or: pip install dekko / pipx install dekko\ndekko --claude-install # add the /map command + MCP server to Claude Code, then restart\n```\n\nExtras (`dekko[all]` for ~55 more languages, `dekko[search]` for\nembedding search), installing from a local clone, and uninstalling are\nin [docs/install.md](docs/install.md).\n\n## Quick start\n\n```sh\ncd my-project\ndekko map # writes .dekko/MAP.md + .dekko/map.json\ndekko summary # ~40-line digest: dirs, hotspots, entry points\n```\n\n`.dekko/` is git-ignored by default; the map regenerates on demand, so\nyou rarely need to run `dekko map` again by hand.\n\n## Documentation\n\n- [docs/install.md](docs/install.md) — installation, extras, local\n clone, uninstall\n- [docs/cli.md](docs/cli.md) — every CLI command, symbol targets,\n excluding files, notes, daemon mode, language support\n- [docs/claude-code.md](docs/claude-code.md) — the `/map` plugin, push\n hooks, Claude Code skills, the MCP server, and Cline\n\n## Learn more\n\n- [CHANGELOG.md](CHANGELOG.md) — per-version history\n- [CONTRIBUTING.md](CONTRIBUTING.md) — dev setup, testing, releasing\n- [benchmarks/](benchmarks/) — token-efficiency measurements, including\n a 7-repo real-world comparison against a plain Read/Grep workflow\n", "bytes": 5751, "sha": "c39cd0c17247998e33f031332e90bde5159a64d831fb2b37d536a4861beee1de", "repo_slug": "aahlijia/dekko", "fonte": "repo", "truncated": false, "api": "https://agentalog.com/api/listings/mcp_io_github_aahlijia_dekko_37a2a058/readme" }