Skip to content
Back to skills

Codebase Memory

ASecurity

Structural code intelligence — query a tree-sitter knowledge graph of a codebase instead of grepping file-by-file. Use for who-calls-what (trace_path), change blast radius (detect_changes), architecture overviews (get_architecture), dead code, and Cypher queries over code structure. Binary at ~/workspace/tools/codebase-memory-mcp/codebase-memory-mcp, CLI mode only.

  • 6 stars
  • 0 votes
  • 0 copies
  • 1 view
  • Added September 30, 2026
ai-agentsrustbashnodegitapi

Works with

  • cli
  • api
  • mcp

Security analysis

A100/100

Scanned September 30, 2026

npx -y skills add aicodedecode/awesome-muse-skills --skill codebase-memory --agent claude-code

Installs into .claude/skills of the current project.

Are you the author of Codebase Memory?

Add the live security badge to your README. It updates with every re-scan.

Security grade badge for Codebase Memory
[![Security: A — Skills Directory](https://www.skillsdirectory.com/api/skills/aicodedecode-codebase-memory/badge)](https://www.skillsdirectory.com/skills/aicodedecode-codebase-memory)

More formats (shields.io, HTML) on the badges page. Keep it an A: scan every change in CI with Pro.

Download with Pro
SKILL.md
---
name: codebase-memory
description: Structural code intelligence — query a tree-sitter knowledge graph of a codebase instead of grepping file-by-file. Use for who-calls-what (trace_path), change blast radius (detect_changes), architecture overviews (get_architecture), dead code, and Cypher queries over code structure. Binary at ~/workspace/tools/codebase-memory-mcp/codebase-memory-mcp, CLI mode only.
---

# codebase-memory (DeusData/codebase-memory-mcp)

Code-intelligence engine: indexes a repo into a persistent knowledge graph
(functions, classes, call chains, HTTP routes) via vendored tree-sitter
grammars, queryable with 17 tools. MIT licensed. 100% local — no API key, no
hosted service, no LLM inside.

Installed 2026-09-30 as **v0.11.0** (release SHA-256 verified against the
published checksums.txt). The project's own installer was deliberately NOT
run: it auto-configures 45 agent client surfaces and writes lifecycle hooks
into agent configs — none of which exist in this runtime, and it violates the
no-installers rule. Binary-only install at:

```
~/workspace/tools/codebase-memory-mcp/codebase-memory-mcp
```

Use **CLI mode only** (`cli <tool> --flags`). Do NOT start the coordination
daemon or the graph UI — both assume a persistent machine; this VM gets
replaced. Do NOT run `install` (see above).

## Indexed projects

| Project | Nodes | Edges | Mode | Indexed |
|---|---|---|---|---|
| raahupsc (`~/workspace/raahupsc`) | 79,666 | 88,310 | fast | 2026-09-30 |

Reindex after large refactors: `cli index_repository --repo-path <path>
--name <name> --mode fast`. (`fast` = structural only; `full` adds local
semantic embeddings — slower, use only if semantic search is needed.)

## Recipes

```bash
CBM=~/workspace/tools/codebase-memory-mcp/codebase-memory-mcp

# Who calls a function (and what it calls) — depth 1-5
$CBM cli --quiet trace_path --function-name findDailyCaInWindow \
  --project raahupsc --direction inbound --depth 2

# Blast radius of the working-tree diff
$CBM cli --quiet detect_changes --project raahupsc

# Codebase overview: languages, routes, hotspots, clusters
$CBM cli --quiet get_architecture --project raahupsc

# Structural / text / semantic search
$CBM cli --quiet search_graph --project raahupsc --label Function \
  --query "publish article"

# Read-only Cypher over the graph (get_graph_schema first)
$CBM cli --quiet get_graph_schema --project raahupsc

# Read a function's source by qualified name
$CBM cli --quiet get_code_snippet --project raahupsc \
  --qualified-name "raahupsc.src.lib.content.getDailyByDateKey"

# Is a path/scope actually indexed and fresh?
$CBM cli --quiet check_index_coverage --project raahupsc --paths '["src/lib/content.ts"]'
```

## Standing cautions

- **Verify against source.** The project's own docs say `check_index_coverage`
  is "not proof of completeness" and to fall back to source for flagged gaps.
  Treat graph answers as leads: confirm in the file before editing.
- **Marketing vs observed:** "average repo in milliseconds" — raahupsc took
  ~60s wall (fast mode). Still one cheap indexing run vs many grep cycles.
- **Right tool for the question:** grep/`muse.read` for text search and small
  files; the graph for *structural* questions (call chains, impact analysis,
  "what breaks if I change X", architecture). Don't reach for it otherwise.
- **Verified 2026-09-30:** `trace_path` on `findDailyCaInWindow` (inbound,
  depth 2) correctly returned 7 callers across page components and wrapper
  functions (`getDailiesInRange`, `getDailyByDateKey`) — where a naive
  `grep -l` surfaced only the definition file. Genuinely better than grep
  for call-chain questions.
- Binary is 299MB (tree-sitter grammars vendored in). Keep it under
  `~/workspace/tools/`; do not copy per-project.

---
## Provenance (system note, 2026-09-30)

- Source: https://github.com/DeusData/codebase-memory-mcp, v0.11.0,
  tag 2026-09-15. License: MIT. Author publishes SLSA-3 provenance,
  OpenSSF Scorecard, and per-release VirusTotal scans; this install
  verified the tarball SHA-256 against release checksums.txt only.
- Chosen over MCP-server registration: this runtime has no MCP client
  surface, so CLI mode is the integration.

Attribution

Is this your skill, or is something wrong with this listing? Request removal or report an issue. Author removals are honored within 72 hours.

Comments

Loading comments…