Player not loading? Watch on YouTube
This tutorial explains how to set up Codebase Memory MCP for a coding assistant and index a repository into a persistent structural knowledge graph. The speaker describes it as a local standalone executable that needs neither Docker nor a background language runtime, and claims graph queries reduce raw file reading and token use.
The walkthrough covers a Bash installer for macOS and Linux, with a UI flag for the visualizer. Windows uses a PowerShell sequence that includes unblocking the downloaded script before installation. The speaker discusses possible Microsoft Defender false positives and recommends checking the SHA-256 digest against official GitHub releases before adding a folder exclusion.
The installer detects supported agents and adds MCP configuration. After restarting the IDE or terminal, users prompt the agent to index the project. The speaker warns that a lowercase Windows drive letter can cause the background cleaner to delete the newly built database.
The browser visualizer runs at localhost:9749 and groups connected code into architectural clusters. The tutorial also covers a memory limit for graph construction and disabling the Git-based watcher. It closes with bidirectional call graph tracing for impact analysis, using a core utility function and its upstream API routes as the example. These are the walkthrough's stated capabilities, rather than independently measured performance results.
