Codebase Memory MCP — Code Knowledge Graph Server

What it is

Codebase Memory MCP is a high-performance MCP server that indexes your codebase into a persistent knowledge graph. It supports 66 programming languages, provides sub-millisecond queries, and reduces the tokens needed for code context by replacing raw file content with structured graph lookups. The server runs locally and integrates with AI assistants through the Model Context Protocol.

The project targets developers using AI coding assistants who want better code understanding without sending entire files as context. By pre-indexing the codebase into a graph of functions, classes, imports, and dependencies, the MCP server provides precise answers to structural queries.

How it saves time or tokens

Sending raw files to an AI assistant wastes tokens on irrelevant code. Codebase Memory MCP indexes your codebase once and serves targeted queries: 'what calls this function,' 'what does this class depend on,' 'show me the type hierarchy.' Each query returns only the relevant graph nodes, potentially using 99% fewer tokens than raw file context. The persistent index survives restarts, so re-indexing only processes changed files.

How to use

1. Add to your MCP configuration:

{
  "mcpServers": {
    "codebase-memory": {
      "command": "npx",
      "args": ["-y", "codebase-memory-mcp"]
    }
  }
}

1. Restart Claude Code. The server indexes your codebase automatically on first use.

1. Ask Claude structural questions about your codebase, and it will use the knowledge graph for precise answers.

Example

Queries the knowledge graph can answer efficiently:

# Instead of reading entire files, the MCP server returns graph data:

> What functions call processPayment()?
  - checkout.ts:handleCheckout (line 45)
  - subscription.ts:renewSubscription (line 112)
  - admin.ts:manualCharge (line 78)

> What are the dependencies of the AuthService class?
  - UserRepository (import from ./repositories)
  - TokenManager (import from ./utils)
  - Logger (import from @/shared)

> Show the type hierarchy for BaseController
  - BaseController
    - UserController
    - ProductController
    - OrderController

Related on TokRepo

• MCP GitHub Integration — GitHub access through MCP for code repositories
• AI Memory Providers — persistent memory solutions for AI agents

Common pitfalls

• Initial indexing of large codebases (100K+ files) can take several minutes; subsequent incremental updates are faster
• The knowledge graph is local to your machine; team members each maintain their own index
• Language support varies in depth; popular languages (TypeScript, Python, Go) have richer graph structures than niche languages