@uworx/code-memory

Local-first repository intelligence engine exposed via the Model Context Protocol (MCP). Build a persistent semantic memory layer over any codebase.

code-memory transforms repositories into queryable intelligence — extracting symbols, relationships, and semantic understanding — and exposing it through MCP tools designed for AI coding agents.

Installation

From your terminal run:

npx -y @uworx/code-memory --version

It should print the installed version and exit. You can then proceed to configure your MCP client (VS Code, Cursor, Claude Desktop, etc.) to use the code-memory MCP server as described below.

Configuration

Configure the MCP server in your client (VS Code, Cursor, Claude Desktop, etc.) by adding the following to your MCP settings:

{
  "mcpServers": {
    "code-memory": {
      "command": "npx",
      "args": ["-y", "@uworx/code-memory"]
    }
  }
}

This indexes the current working directory. To index a different folder, pass --repo (each argument as a separate array element):

{
  "mcpServers": {
    "code-memory": {
      "command": "npx",
      "args": ["-y", "@uworx/code-memory", "--repo", "/path/to/your/project"]
    }
  }
}

How It Works

The @uworx/code-memory package is a lightweight CLI wrapper. On npm install, it downloads the platform-specific native binary from GitHub Releases. The binary is a self-contained .NET single-file publish with no runtime dependencies.

MCP Tools

| Tool | What it gives you | |---|---| | ping | Health check + indexing status (indexingCompleted, fileWatcherActive) | | semantic_search | Natural language code search with optional similarity threshold | | trace_dependency | Symbol dependency tracing (upstream/downstream/both, configurable depth) | | get_architecture_overview | Component structure, language breakdown, file/symbol counts | | get_edit_context | Source code, dependency chains, related symbols, and test coverage for a symbol | | find_related_code | Related symbols via dependency graph (breadth-first, filterable by type) | | impact_analysis | Change impact — downstream dependencies, affected files, components, test coverage | | get_component_clusters | Logical groupings based on inter-component coupling density | | get_symbol_history | Git commit history for a symbol (commits, authors, dates) | | get_hotspots | Most frequently changed files ranked by commit count | | sql_query | SQL queries over indexed data — SELECT/WHERE/ORDER BY/GROUP BY/HAVING, CTEs, derived tables, aggregates, vector search via ORDER BY Similarity DESC | | rescan_repository | Trigger full re-index (clear all data, re-scan, re-store) | | get_repository_root | Returns the root path of the currently active repository |

All tools return structured JSON.

Requirements

• Node.js 18+ (for npx / npm install)
• No .NET runtime required — the binary is self-contained

Supported Platforms

• Windows: x64
• Linux: x64
• macOS: coming soon

Indexing Note

Indexing is non-blocking. Poll the ping tool until indexingCompleted is true before calling other tools, or results may be empty/partial.

.codememory Folder

CodeMemory creates a .codememory folder in the target repository root to store index data, logs (Log.*.txt), and — in future versions — additional cached artifacts. This folder is managed entirely by CodeMemory and is kept clean. You can safely add .codememory/ to your project's .gitignore; the MCP server handles cleanup internally.

Learn More

Full documentation: github.com/khurram-uworx/CodeMemory

License

Apache-2.0