Scaffold

Name: Scaffold
Author: Beer-Bears

@Beer-Bears

关于 Scaffold

Scaffold is a specialized RAG system that transforms entire codebase into a living knowledge graph, capturing the deep structural relationships between files, classes, and functions. This provides AI agents with the crucial context to accurately understand, maintain, and reason a

#graph-rag #rag #llm #knowledge-graph #code-analysis #generative-ai #context-retrieval

基本信息

分类

开发工具

传输方式

stdio

发布者

Beer-Bears

提交者

Ketrin XD

配置

使用下面的配置,将此服务器添加到你的 MCP 客户端。

{
  "mcpServers": {
    "Scaffold": {
      "command": "docker",
      "args": [
        "run",
        "--rm",
        "--name",
        "scaffold-mcp",
        "-p",
        "8000:8080",
        "-v",
        "${PROJECT_PATH}:/app/codebase",
        "-e",
        "CHROMA_SERVER_HOST=${CHROMA_SERVER_HOST}",
        "-e",
        "CHROMA_SERVER_PORT=${CHROMA_SERVER_PORT}",
        "-e",
        "CHROMA_COLLECTION_NAME=${CHROMA_COLLECTION_NAME}",
        "-e",
        "NEO4J_USER=${NEO4J_USER}",
        "-e",
        "NEO4J_PASSWORD=${NEO4J_PASSWORD}",
        "-e",
        "NEO4J_URI=${NEO4J_URI}",
        "-t",
        "ghcr.io/beer-bears/scaffold:latest"
      ],
      "env": {
        "CHROMA_SERVER_HOST": "<CHROMA_HOST>",
        "CHROMA_SERVER_PORT": "<CHROMA_PORT>",
        "CHROMA_COLLECTION_NAME": "scaffold_data",
        "NEO4J_USER": "<NEO4J_USER>",
        "NEO4J_PASSWORD": "<NEO4J_PASSWORD>",
        "NEO4J_URI": "bolt://<NEO4J_USER>:<NEO4J_PASSWORD>@<NEO4J_HOST>:NEO4J_PORT>",
        "PROJECT_PATH": "<PATH_TO_YOUR_PROJECT>"
      }
    }
  }
}

工具

未检测到工具

工具是从 README 中自动提取的。维护者可以在 ## Tools 标题下列出工具,即可填充这部分内容。

概览

What is Scaffold?

Scaffold is a specialized Retrieval-Augmented Generation (RAG) system that provides deep, structural understanding of large codebases to LLMs and AI agents. It transforms source code into a living knowledge graph, enabling precise, context-aware interactions. This MCP server allows clients like Cursor to interface with the Scaffold knowledge graph to construct, maintain, and reason about complex software projects.

How to use Scaffold?

First, run the required database dependencies (Neo4j and ChromaDB) in Docker on a shared network. Then, configure your MCP client (e.g., mcp.json in Cursor) with a Docker command and environment variables that point to your databases and the absolute path to your codebase. The server exposes a single tool get_code_entity_information that can be invoked from the AI agent.

Key features of Scaffold

Graph-Based RAG: Represents codebases as interconnected graphs of files, classes, and functions.
Living Knowledge Base: Keeps the AI's understanding synchronized with code structure.
Eliminate Context Blindness: Provides project-specific architecture and logic to agents.
Precise Context Injection: Delivers targeted information about code entities.
Hybrid Search: Combines graph traversal with vector search for comprehensive context.

Use cases of Scaffold

Ask an AI agent what a specific class (e.g., PaymentService) does.
Find the source code and relationships of a particular function (e.g., save_graph_to_db).
Understand what a code entity like FileNode is and which other methods use it.

FAQ from Scaffold

What are the runtime dependencies for Scaffold?

Scaffold requires Neo4j and ChromaDB to be running in Docker on a shared network before the MCP server container starts.

How do I fix a "Connection Refused" error?

Ensure the database containers (scaffold-neo4j and scaffold-chromadb) are running and all three containers are on the same Docker network. Also verify that the user, password, and hostnames in the environment variables match the running databases.

Why does the server report "File not found"?

The PROJECT_PATH environment variable must be an absolute path on your host machine. Relative paths (like . or ~/) may not work correctly.

Where does Scaffold store its data?

Data is stored in the Neo4j graph database and ChromaDB vector store, both running in Docker containers with persistent volumes (neo4j_data and chroma_data).