Case Study

Our delivery pipeline is Claude-native.

How PineApps rebuilt its Dynamics 365 delivery practice around Claude Code, agents and the Model Context Protocol — and what that pipeline ships.

The starting point

Quality that depends on people doesn't scale.

PineApps delivers business applications on Dynamics 365, Power Platform, Power BI and Azure for 15+ mid-market companies in Turkey, across manufacturing, machinery and financial services.

In a boutique consultancy, delivery quality traditionally depends on who happens to be on the project. We wanted the engineering practice itself to be the asset: the same analysis discipline, the same schema standards, the same review checklist — on every project, at every customer. So in 2025 we rebuilt the practice around Claude.

The result is the one metric that matters: a team this size serves 15+ customers, because the pipeline — not headcount — carries the practice.

What we built

Standards as code, operations by agents.

1

Delivery standards packaged as Claude skills

30+ company standards — requirements analysis, Dataverse schema design, plugin development, code review, testing, customer communication — live as Claude skills in one monorepo. Every Claude session on every project loads the same playbook, which makes the practice reproducible rather than dependent on individuals.

2

Our own MCP server for Dataverse

We built a Model Context Protocol server for Microsoft Dataverse: ~30 tools covering tables, columns, relationships, option sets, solutions and records. Schema work that used to be hours of clicking through the maker portal is executed by Claude agents against live customer environments — connection-aware, solution-scoped, and logged.

3

A role-based, multi-agent lifecycle

Claude Code drives the full delivery lifecycle, from analysis and data-model design through plugin development, code review, testing and documentation, with role-based agents following those skills at every step.

4

Guardrails that make agent deploys safe

Nothing reaches a live environment unaccounted for. Deploy tooling refuses uncommitted, unmerged or out-of-sync sources; every deployed artifact is stamped with the commit it came from; and a weekly drift audit compares what is live against the repository. What is live is what is in main — by construction. And every deploy is a deliberate, engineer-triggered act: agents prepare and verify; a person ships.

Inside the pipeline

One request, end to end.

A typical change request at one of our Dynamics 365 customers now flows like this:

1

The request lands and the requirements-analysis skill turns it into a functional spec with a data model — the same questions asked, the same structure produced, every time.

2

Claude designs the schema to our naming standards and creates the tables, columns and relationships in the live Dataverse org through our MCP server — no maker-portal clicking.

3

Form scripts and plugins are written to company standards and pass through the code-review skill before anything ships.

4

Guarded deploy tooling pushes the artifacts — refusing anything uncommitted or unmerged, and stamping every deploy with its commit.

5

Documentation and the customer-facing delivery note come out of the same session, in the customer's language, following the business-communication skill.

The loop

From CRM case to CRM resolution.

The pipeline is a closed loop, with the customer's own system at both ends. A support case in Dynamics 365 becomes a GitHub issue; Claude Code picks it up in VS Code; agents deliver through MCP and guarded deploys; and the resolution — time entry included — flows back into Dynamics 365.

The glue of the loop — the D365↔GitHub integration, the Azure Functions behind it, the portal where cases arrive — was itself built with Claude.

Dynamics 365

Customer opens a support case

GitHub

Case becomes an issue; all work tracked on main

Claude Code

Agents + skills do the work in VS Code

Guarded deploy

MCP schema ops into the live environment

Dynamics 365

Resolution + time entry flow back to the case

What it ships

The proof is the output.

This isn't an internal experiment. The same pipeline ships customer-facing work — including the page you are reading.

Live

pineapps.co

This website

Designed and built end-to-end with Claude Code — brand system, copy, code and deployment to Azure Static Web Apps. What you are looking at is pipeline output.

Live

Dynamics 365 delivery

The day job

Schema, plugins, form scripts and reports for 15+ Dynamics 365 customers ship through this pipeline every week — with agents executing against live environments.

Live

PineHR

HR self-service PWA

Leave requests, expenses, approvals and announcements in an employee self-service PWA on a complete Dataverse HR data model — designed, built and shipped through this pipeline.

Live

Spectra

AI website analysis

A Claude analysis pipeline that turns any website into scored, prioritised recommendations across technical health, content, SEO and UX — publicly accessible, free to try.

Launching Sept 2026

PineLead

AI lead capture

Business-card photo to CRM lead in seconds, with Claude doing the extraction. Launches at MAKTEK Istanbul with eight exhibitor customers in machinery manufacturing.

Live

Customer Portal

Support portal

The front door of the loop above: a branded portal where customers open and track their support cases, connected live to Dynamics 365 — built with Claude, in daily use at our customers.

The broader product line — comPass, ColorGrid, PineDocs, BriefCast — is built and maintained through the same pipeline. See Products.

Why it matters

An entry point the market is asking for.

Our installed base of Dynamics 365 and Power Platform customers gives Claude an obvious entry point: agentic workflows on top of existing CRM, ERP and reporting data — including Logo ERP integration that is specific to the Turkish market.

And the demand is already at our door: customers who see Claude in our delivery work ask how to adopt it in their own teams. That is the practice we are building next.

Want this pipeline pointed at your operation?

Tell us where the manual work is. We'll show you what an agentic workflow on your own CRM, ERP and reporting data looks like.

Talk to us