Github-Assistant-MCP-Server

An MCP Server written using Java and Spring AI which exposes tools to allow common AI agents to perform various Github related tasks.

Introduction • Architecture • Setup • Usage

Introduction

Model Context Protocol (MCP) is a standard open-source protocol used to allow common AI agents to interact with external tools in a standard manner.

This project is an implementation of an MCP Server written using Spring AI that exposes tools to allow common AI agents such as Copilot, to perform various Github related tasks. Generally, these AI agents aren't able to access repositories, especially private repositories, and perform actions such as listing issues, listing PR's etc. This server allows these agents to perform such actions by exposing tools that they can call.

The tools currently exposed by the MCP server include:

• list_issues_limited: List a specific number of issues in a given repository for a given owner.
• list_issues_default: List a 5 issues in a given repository for a given owner.
• get_issue: Get details about a specific issue.
• list_pull_requests: List general details about all PR's (open PR's for now).
• get_pull_request: Get details about a specific PR.

The philosophy behind designing the MCP tools is : "Agent-friendly, not API-shaped"

Architecture

Setup

Setting up Github Token

Corresponding to github.token=${GITHUB_TOKEN} in application.properties, an environment variable named GITHUB_TOKEN will have to be set. This should be a valid Github Personal Access Token (PAT) that has the necessary permissions to perform the desired operations on Github. This token will be used by the MCP server to authenticate API requests to Github.

Adding MCP Server to Github Copilot

Go to Tools -> Github Copilot -> Edit Settings... and click on "Model Context Protocol (MCP)" and click on "Configure" to open mcp.json.

Add the below to this json file:

"github-assistant": {
  "type": "streamable-http", 
  "url": "http://localhost:8080/mcp"
}

For Developers

Run the below commands in local before pushing code to remote.

Code formatting

mvnw spotless:check

If there are formatting issues, run the below command to fix them.

mvnw spotless:apply

Static analysis

mvnw spotbugs:check

Style/complexity analysis

mvnw pmd:check

Unit tests and Code coverage

Always ensure to add or update unit tests for any new code/features and run all tests before pushing to remote.

Run the below command to execute unit tests and generate code coverage report (for local analysis, if

... View full README on GitHub