DEV Community: James Murrell

The Business Analyst's Guide to AI-Assisted Tool Development

James Murrell — Wed, 07 Jan 2026 22:05:09 +0000

I built a few production tools in 3 months without being a developer. Total time saved per week: 8 hours. Here's exactly how I did it.

Three months ago I was doing this manually:
4 hours/week: Accessibility compliance checking
2 hours/week: Care provider research across multiple regulators
1.5 hours/week: Preprod environment testing
30 mins/week: Rewriting documentation for different audiences

Now I have tools that do it:
🐕 The Moby Accessibility Checker → 30 seconds
🏥 Care Provider Finder → 2 minutes
🔍 MGC Web Page Analyzer → 5 minutes
✍️ Prompt Refiner → Instant

Time saved: ~8 hours per week

I'm 57. Business Analyst . Not a developer.
But I learned how to work effectively with AI to build tools I actually use.
What I built:

The Moby Accessibility Checker
Care Provider Finder
Web Page Analyzer v2.0
Prompt Refiner

All on GitHub. All working. All used daily.

How this happened: Not overnight. Not magic. Proper work:
15-25 hours per tool
Constant testing against real workflows
Multiple rewrites when things didn't work
Real problems solved, not theoretical exercises

What made it possible:
I brought domain expertise. AI brought code generation.
I knew what needed to be built. AI knew how to build it.
I tested everything. AI iterated until it worked.
That partnership is the key.

What I learned:
✅ Start with one painful, repetitive problem
✅ Simple working tools beat complex broken ones
✅ Your BA skills translate perfectly (requirements, testing, iteration)
✅ Documentation matters more than elegant code
✅ Domain knowledge is more valuable than coding skills
❌ Token limits will surprise you (had to refactor entire approach)
❌ Not all data is available (still frustrated about Wales)
❌ First version is never right
❌ Screenshots don't scale (learned that the hard way)

I wrote it all down.
The process. The decisions. The mistakes. The code.
"The Business Analyst's Guide to AI-Assisted Tool Development". Real examples with working code. Real problems and how I solved them. Real failures and how I recovered.

No theory. No hype. Just the practical process that worked.
£5. Launches this week.

hashtag#update Get it here: https://lnkd.in/efVeXn6F

If you're spending hours on repetitive work and wondering "could I build something to fix this?" - this shows you exactly how.

Questions welcome. Still learning myself.

From Problem to Solution: Creating Prompt Refiner

James Murrell — Mon, 08 Dec 2025 21:35:15 +0000

Building Prompt Refiner: An MCP Extension for Better AI Prompts

The Problem

As a Business Analyst, I use Claude daily for requirements documentation, user stories, and stakeholder communications. But I found myself constantly rewriting prompts to get better results.

The cycle was frustrating:

Write a prompt
Get okay results
Rewrite to be more specific
Get better results
Realize I should have been more concise
Rewrite again...

I needed a faster way.

The Solution

I built Prompt Refiner - an MCP extension that instantly shows you 4 refined versions of any prompt:

Concise - Direct and to the point
Detailed - Rich with context
Creative - Encourages imagination
Analytical - Structured and logical

How It Works

Ask Claude to refine your prompt
See 4 variants instantly in a React artifact
Click copy on your preferred variant
Use it immediately

Technical Implementation

Built with:

Node.js MCP server
SQLite for history tracking
React artifacts for UI
Configurable storage system

Key Features

1. Configurable Storage

Perfect for enterprise environments. Store data on network drives, local folders, or in-memory.

2. Graceful Fallback

Works even on locked-down machines with automatic fallback.

3. History Tracking

All refinements saved for future reference.

Results

5-minute setup
Zero external dependencies
Works entirely in Claude Desktop
Open source (MIT license)

Try It Yourself

GitHub: https://github.com/cs97jjm3/prompt-refiner

Conclusion

If you use AI in your workflow, prompt engineering is a skill worth developing. Prompt Refiner is a tool to help you get there faster.

What prompt engineering tools do you use? Share in the comments!

How I Built an MCP Server to Automate My BA Workflow

James Murrell — Sat, 06 Dec 2025 15:51:56 +0000

Introduction

Every sprint planning session, same questions:

"When will we ship?"
"What fits in this sprint?"
"Does this prioritisation actually make sense?"

I'm a Business Analyst. I answer these questions weekly. They're not hard, just repetitive.
So I built tools to answer them automatically.

What is MCP?

Model Context Protocol lets you extend what Claude can do. Instead of just chatting, you give it tools - functions it can call to get real answers.
Think of it like plugins for AI.

*The Problem I Was Solving
*
BA work involves lots of small calculations:

Working days between dates (excluding bank holidays)
Release dates based on velocity
Whether priorities actually make sense
Formatting user stories consistently

Not complex. But time-consuming when you do them repeatedly.

*What I Built
*
BA Workflow Tools - 17 MCP tools in one server:
Sprint & Release Planning

calculate_working_days - excludes weekends and UK bank holidays
calculate_sprint_dates - generate multiple sprints at once
calculate_release_date - story points + velocity = ship date
calculate_velocity - with capacity adjustments for holidays

MoSCoW Prioritisation

calculate_moscow_priority - breakdown with percentages
plan_moscow_capacity - what fits in your sprint
validate_moscow_dependencies - catch problems before they happen

User Stories

format_user_story - proper structure with acceptance criteria

Plus: timezone tools, estimation converters, text utilities

Code Walkthrough

javascript// Example: Release date calculator
{
  name: "calculate_release_date",
  description: "Calculate estimated release date based on story points remaining, team velocity, and sprint length",
  inputSchema: {
    type: "object",
    properties: {
      storyPointsRemaining: { type: "number" },
      teamVelocity: { type: "number" },
      sprintLength: { type: "number" },
      startDate: { type: "string" }
    },
    required: ["storyPointsRemaining", "teamVelocity", "sprintLength", "startDate"]
  }
}

The logic: divide points by velocity, multiply by sprint length in weeks, add working days to start date. Simple - but now Claude does it for me.

How to Install

Clone the repo
npm install
Add to Claude Desktop config
Restart Claude Desktop

Real Examples

Me: "I've got 85 story points left, velocity is 25 per sprint, sprints are 2 weeks. When will we finish?"
Claude: [calls calculate_release_date] "Based on your velocity, you'll need approximately 4 sprints. Estimated completion: March 14th 2025."
Me: "Check if any of my Must-haves depend on Won't items"
Claude: [calls validate_moscow_dependencies] "Found 2 issues: REQ-045 (Must) depends on REQ-089 (Won't)..."

What I Learned Building MCP Servers

This is my third MCP server. The other two are at work:

Date Operations - working days across 14 countries
Web Analyser with Login - page analysis behind authentication

Key lessons:

Start with problems you solve repeatedly
Keep tools focused - one job per tool
Good descriptions matter - Claude uses them to decide when to call your tool
Test with real prompts, not just unit tests

What's Next

Planning to add:

RAID log management
Stakeholder matrix tools
More estimation converters

PRs welcome.

Links

GitHub: https://github.com/cs97jjm3/ba-workflow-tools
MCP Documentation: [link]
Follow me for more BA automation content