How do I install the Codebase Memory skill?

Run npx skillmds@latest add dreamlab-ai/codebase-memory in your terminal (requires Node.js), paste this page's agent-chat prompt into Claude, Cursor, or any MCP-connected agent, or download the SKILL.md file and copy it into your agent's skills directory.

Is Codebase Memory safe to use?

This skill has not completed SkillMD's automated safety review yet. SkillMD never runs a skill's scripts for you; review the SKILL.md before installing.

Which AI agents work with Codebase Memory?

This skill is tagged as working with Claude Code, Claude.ai, OpenAI Codex. SKILL.md is an open format, so most agents that read a skills directory can load it too.

Is Codebase Memory free to use?

Yes. Installing skills from SkillMD is free, and the skill stays under its author's original license.

Who published Codebase Memory?

dreamlab-ai (@dreamlab-ai) published this skill. Their other Agent Skills are listed on their SkillMD profile.

Codebase Memory by dreamlab-ai — Agent Skill

Name: Codebase Memory
Author: dreamlab-ai

Codebase Memory

Structural code-intelligence MCP for large codebases — trace call graphs, get architecture overviews, score git-diff impact, search symbols, and manage ADRs from a persistent tree-sitter knowledge graph. Use on large or unfamiliar repos (500+ files) for "what calls X?", blast-radius / diff-risk analysis, module architecture, or symbol lookup — when Grep/Glob are slow or noisy. Prefer direct Read/Grep for small projects and one-off fixes.

dreamlab-ai cd02e02 2 files · 8.5 KB Updated Sep 17, 2026

File contents

Codebase Memory — Structural Code Intelligence MCP

Builds a persistent knowledge graph of your codebase and answers structural questions (call chains, architecture, diff impact) without pulling file contents into context. Single binary, no API keys, tree-sitter across 66 languages. The binary is pre-installed in this container; the MCP server starts on demand.

When to Use

Deploy for large codebases (500+ files) when:

Call chains: "What calls ProcessOrder? Show the full chain."
Architecture: "Give me an overview of the payment module."
Diff impact: "What does this git diff affect? Risk score it."
Symbol search: "Find all classes that implement IRepository."
ADR management: "Create an ADR for switching to PostgreSQL."
Session start: index once, then query instantly for the rest of the session.

When NOT to

Small projects (< 100 files) — direct Read/Grep is sufficient.
One-off fixes — the indexing overhead (30s–5min) isn't worth it.
Already indexed — check index_status before re-indexing.

Quick Path

# 1. Index once (~30s small repos, up to ~5min for 1M+ files)
index_repository(repo_path="/home/devuser/workspace/project")

# 2. Query the graph instead of Grep/Glob
trace_call_path(...)   # who calls / is called by X (depth 1–5)
get_architecture(...)  # languages, routes, endpoints, clusters
detect_changes(...)    # git diff → affected symbols + risk score
search_graph(...)      # find a class/symbol by name
get_code_snippet(...)  # fetch source by qualified name

# 3. Verify freshness at session start
index_status(...)      # if stale (last sync > latest commit) → index_repository

Full tool catalog (14 tools), parameters, and the token-efficiency figures are in references/reference.md.

Permanent Project Upgrade (one-time setup, lasting benefit)

Once a project is indexed, append a "Codebase Memory MCP (ACTIVE — USE FIRST)" block to its CLAUDE.md / CLAUDE.local.md so future sessions default to these tools before Grep/Glob. Consider this when starting exploratory work on an unfamiliar large repo, before a refactor that needs blast-radius awareness, or when "what calls X?" questions recur. The ready-to-paste block, its tool-routing table, and the trigger checklist are in references/reference.md.

Tool catalog, token-efficiency table, architecture, env vars, setup, and cross-skill integration: references/reference.md.
Attribution: codebase-memory-mcp by DeusData (https://github.com/DeusData/codebase-memory-mcp).

---name: codebase-memorydescription: Structural code-intelligence MCP for large codebases — trace call graphs, get architecture overviews, score git-diff impact, search symbols, and manage ADRs from a persistent tree-sitter knowledge graph. Use on large or unfamiliar repos (500+ files) for "what calls X?", blast-radius / diff-risk analysis, module architecture, or symbol lookup — when Grep/Glob are slow or noisy. Prefer direct Read/Grep for small projects and one-off fixes.---# Codebase Memory — Structural Code Intelligence MCPBuilds a persistent knowledge graph of your codebase and answers structuralquestions (call chains, architecture, diff impact) without pulling file contentsinto context. Single binary, no API keys, tree-sitter across 66 languages. Thebinary is pre-installed in this container; the MCP server starts on demand.## When to UseDeploy for large codebases (500+ files) when:- **Call chains**: "What calls ProcessOrder? Show the full chain."- **Architecture**: "Give me an overview of the payment module."- **Diff impact**: "What does this git diff affect? Risk score it."- **Symbol search**: "Find all classes that implement IRepository."- **ADR management**: "Create an ADR for switching to PostgreSQL."- **Session start**: index once, then query instantly for the rest of the session.### When NOT to- Small projects (< 100 files) — direct Read/Grep is sufficient.- One-off fixes — the indexing overhead (30s–5min) isn't worth it.- Already indexed — check `index_status` before re-indexing.## Quick Path```# 1. Index once (~30s small repos, up to ~5min for 1M+ files)index_repository(repo_path="/home/devuser/workspace/project")# 2. Query the graph instead of Grep/Globtrace_call_path(...) # who calls / is called by X (depth 1–5)get_architecture(...) # languages, routes, endpoints, clustersdetect_changes(...) # git diff → affected symbols + risk scoresearch_graph(...) # find a class/symbol by nameget_code_snippet(...) # fetch source by qualified name# 3. Verify freshness at session startindex_status(...) # if stale (last sync > latest commit) → index_repository```Full tool catalog (14 tools), parameters, and the token-efficiency figures are in[`references/reference.md`](references/reference.md).## Permanent Project Upgrade (one-time setup, lasting benefit)Once a project is indexed, append a "Codebase Memory MCP (ACTIVE — USE FIRST)"block to its `CLAUDE.md` / `CLAUDE.local.md` so future sessions default to thesetools before Grep/Glob. Consider this when starting exploratory work on anunfamiliar large repo, before a refactor that needs blast-radius awareness, orwhen "what calls X?" questions recur. The ready-to-paste block, its tool-routingtable, and the trigger checklist are in[`references/reference.md`](references/reference.md#permanent-claudemd-upgrade-block).## More- Tool catalog, token-efficiency table, architecture, env vars, setup, and cross-skill integration: [`references/reference.md`](references/reference.md).- Attribution: codebase-memory-mcp by DeusData (https://github.com/DeusData/codebase-memory-mcp).

Frequently asked questions