Engineering Intelligence MCP

A production-oriented Model Context Protocol (MCP) server for Express.js applications that enables AI clients to inspect backend systems through structured engineering tools instead of direct file system access.

The server exposes backend engineering context including Git history, Winston logs, OpenAPI/Swagger contracts, and sandboxed source code through MCP tools. The implementation has been validated using both the official MCP Inspector and Claude Desktop.

Built on top of https://github.com/sanchit0496/vectordb-rag

Features

• MCP-compliant server using the official MCP SDK

• Tool Registry based architecture

• Git history provider

• Winston log provider

• Dynamic Swagger/OpenAPI provider

• Sandboxed source code provider

• Compatible with Claude Desktop

• Validated using the official MCP Inspector

• Extensible provider architecture

• Read-only engineering context

Architecture

flowchart TD

Client[Claude Desktop / MCP Inspector]

Transport[STDIO Transport]

Server[MCP Server]

Registry[Tool Registry]

Git[Git Provider]

Swagger[Swagger Provider]

Logs[Winston Provider]

Source[Source Code Provider]

Express[(Express Application)]

Client <--> Transport
Transport <--> Server
Server --> Registry
Registry --> Git
Registry --> Swagger
Registry --> Logs
Registry --> Source

Git --> Express
Swagger --> Express
Logs --> Express
Source --> Express

Provider Responsibilities

Tech Stack

Installation

git clone <repository>

npm install

npm run dev

Configure Claude Desktop:

{
  "mcpServers": {
    "engineering-intelligence": {
      "command": "node",
      "args": [
        "/absolute/path/mcp/server.js"
      ]
    }
  }
}

Restart Claude Desktop.

Available MCP Tools

Working Example

Related MCP server: runtime-mcp-connect

Prompt

Compare the implementation of POST /orders with its OpenAPI contract.

Tool Execution

Result

The server successfully detected a contract mismatch between the documented API and the controller implementation.

Example:

• Swagger contract expects customerId

• Controller consumes userId

This demonstrates how multiple MCP tools can be orchestrated by an AI client to investigate backend implementation issues.

Testing

The implementation was validated using two independent MCP clients.

MCP Inspector

• Tool discovery

• Tool execution

• Protocol validation

• Response verification

Claude Desktop

Successfully connected without code changes.

Validated:

• Git history inspection

• Winston log retrieval

• Swagger analysis

• Source code inspection

This confirms interoperability with real MCP clients.

Project Structure

project/
├── app.js
├── index.js
├── logger.js
├── swaggerConfig.js
├── controllers/
├── models/
├── routes/
├── validations/
├── config/
├── mcp/
│   ├── server.js
│   ├── registry/
│   │    └── toolRegistry.js
│   └── providers/
│        ├── gitProvider.js
│        ├── logProvider.js
│        ├── swaggerProvider.js
│        └── sourceCodeProvider.js
└── package.json

Roadmap

• AST-based source code analysis

• Sequelize schema provider

• Express route dependency graph

• Architecture summarization

• Middleware analysis

• Configuration provider

• Project dependency analysis

License

Licensed under the ISC License.