Summary

The graph is parsed, not inferred — no LLM calls to build it, so re-indexing is cheap enough to run continuously.

That choice is why the impact and evolution questions are answerable at all: every symbol carries its version history, so "what changed and what does it touch" is a graph query rather than a diff you read. Cross-service API topology only appears once you have indexed both ends, which is worth doing before you ask about it.

What it is

A local code knowledge graph for AI coding agents: symbols as nodes, calls and imports as edges, with version history attached to every symbol and HTTP topology across services.

What you get

find_symbol and find_code — exact or fuzzy symbol lookup, and hybrid retrieval that handles natural-language queries and string literals baked into function bodies
get_symbol_context — callers, callees, community, processes and recent changes for one symbol in a single call
get_impact and find_dependency_path — blast radius before a refactor, and how two symbols relate in the call graph
list_communities, find_central_symbols, find_bridge_symbols and get_codebase_briefing — architecture at a glance for a repo you just met
get_evolution, get_timeline, get_changes_since, detect_changes and get_cochange_context — what changed, when, and what historically changes alongside it
find_api_endpoints, find_api_calls, get_api_topology and get_service_diagram — HTTP endpoints and cross-service call graphs across repos you have indexed
find_dead_code, calculate_cyclomatic_complexity and find_most_complex_functions
index_directory, watch_directory and check_job_status — indexing runs async, and watched paths make in-progress edits queryable
get_source_window — a span of a file with surrounding context, cheaper than reading the whole file

Requirements

Npm install -g memtrace, which sets up the binary, skills and MCP config for Claude Code, Claude Desktop, Cursor, Codex, Windsurf, VS Code, Hermes, OpenCode and Kiro. Memtrace is in private beta — you need access before the CLI is useful. Indexing is local and needs no API key; the graph lives in a per-repo .memdb directory.

Setup effort

One command plus a key — npm install -g memtrace, then supply credentials