What is MCP?

The Problem: AI Can Think, But Can't Do

AI models like Claude, GPT, and Gemini are incredibly capable at reasoning and generating text. But by themselves, they can't:

• Read files on your computer
• Run terminal commands
• Search your codebase
• Make API calls
• Coordinate with other agents

They're brilliant minds trapped in a text box.

The Solution: MCP (Model Context Protocol)

MCP is an open protocol created by Anthropic that lets AI models call external tools. Think of it like giving the AI a set of function buttons it can press.

1. AI decides it needs to read a file
2. AI calls: read_file(path="src/app.ts")
3. MCP server executes the request
4. Result returns to AI
5. AI continues with the information

MCP servers expose tools (functions the AI can call) and resources (data the AI can read). Any application that speaks the MCP protocol can connect to these servers.

MCP Clients (Who Uses MCP)

These applications have built-in MCP support, meaning they can connect to MCP servers:

• Claude Code - Anthropic's CLI for coding with Claude
• Claude Desktop - The desktop app
• Cursor - AI-native code editor
• Continue - VS Code extension
• Windsurf - AI coding environment
• Google AntiGravity - Google's agentic coding tool

MCP Servers (What Tools Exist)

MCP servers provide tools for specific purposes. Some examples:

• Filesystem - Read and write files
• Git - Version control operations
• Database - Query databases
• Web - Fetch URLs, search the web
• SCRUM MCP - Multi-agent coordination

The Multi-Agent Problem

Here's where it gets interesting. With MCP, you can now have multiple AI agents working on the same codebase. But this creates new problems:

• Collisions - Two agents editing the same file
• Lost work - Agent A overwrites Agent B's changes
• No communication - Agents don't know what others are doing
• No accountability - Who changed what and why?

How SCRUM MCP Works

SCRUM MCP is an MCP server that provides coordination tools for multi-agent teams. It's like giving your AI agents a shared whiteboard and a file locking system.

The Workflow

1. Declare Intent - Agent announces what files it plans to change
2. Claim Files - Agent locks the files so others can't edit
3. Make Changes - Agent does the work
4. Attach Evidence - Agent proves the work is done (test output, etc.)
5. Release Claims - Agent unlocks files for others

Sprint Collaboration (v0.5)

When multiple agents work on the same task, they can join a Sprint—a shared context space where they:

• Share decisions about architecture
• Publish interfaces they're creating
• Post discoveries about the codebase
• Ask questions and get answers from teammates

Smart Tool Consolidation (v0.5.1)

v0.5.1 reduced MCP tools from 56 to 32 through smart consolidation:

• Action-based tools - scrum_blocker(action: "add|resolve|list") replaces 3 tools
• Type-based tools - scrum_metrics(type: "board|velocity|aging|task") replaces 4 tools
• Profile discovery - scrum_status(profile: "solo|team|full") shows relevant tools
• REST-only admin - WIP limits, gates, templates, webhooks moved to REST API

Compliance Verification (v0.4.0)

SCRUM verifies that agents actually do what they said they would:

• Did they only touch declared files?
• Did they respect boundary files (files they promised not to touch)?
• Did they attach evidence?

If not, claim release is blocked. Trust, but verify.

Getting Started

Ready to try it? Here's how to set up SCRUM MCP:

# Clone and install
git clone https://github.com/tylerbuilds/scrum-mcp
cd scrum-mcp
npm install

# Start the server
npm run dev

# In another terminal, build and run MCP
npm run build
npm run mcp

Then configure your MCP client (Claude Code, Cursor, etc.) to connect to SCRUM. See the IDE Setup Guide for detailed instructions.

Learn More

• MCP Protocol Specification - Official docs
• SCRUM MCP Tool Reference - All 32 tools
• Agent Prompts - Copy-paste prompts for your agents
• Case Study - How SCRUM evolved from v0.1 to v0.5