App Development

Guide to Using Generative AI at GooApps®: Best Practices and MCPs

At GooApps®, the use of generative AI has evolved from experimentation to strict methodology. After more than 1,000 hours of internal research, we have defined a standard based on three pillars: structured prompt engineering (using AI to create its own briefs), model-agnostic selection (Claude for code, GPT for structure), and the adoption of the Model Context Protocol (MCP) to securely connect LLMs with our real tools (Jira, Figma, GitHub).

The core principle: your judgment leads, AI executes

Generative AI speeds up work, but it also accelerates mistakes if not managed wisely. In our experience, errors are rarely the model’s fault; they usually stem from poor context and decisions made without awareness of their real impact.

That’s why at GooApps® we follow an unbreakable rule: if we cannot explain it, test it, and take responsibility for the outcome, we do not use it. This approach aligns directly with our vision of technology, with AI centered on people, ethical judgment, and human accountability, especially in sensitive areas like health and wellness.

The “meta-prompt” technique: stop improvising

Instead of writing prompts manually, we use AI itself to generate robust technical “briefs.”

The structure of a GooApps® engineering prompt:

1. Role: Act as an expert in [Subject].
2. Task: Define the objective in one measurable sentence.
3. Constraints: Output format (JSON, table), length, and style.
4. The “safety clause”: “Before generating anything, ask me all the questions you need to ensure quality. If information is missing, do not invent it.”

This simple pre-validation step reduces hallucinations by 40% and forces the model to reason before writing.

Template 1 — Meta-Prompt to Generate a Robust Prompt

Act as an expert in prompt engineering.
I want you to design a ROBUST PROMPT for the following task:
[TASK]
Before writing the final prompt:
1) Ask all the necessary questions to remove ambiguities (minimum 8).
2) Propose 2 prompt variants: (A) concise and (B) exhaustive.
3) Include a “constraints” section and a “quality criteria” section.
4) Include an example input and an example output.
When you are finished, deliver only the final prompt (version B), ready to copy and paste.

Template 2 — Quality-Enhancing Safety Clause

Before executing the task:
- Ask me all the questions you need to do it properly.
- If information is missing, do not invent it: explain what is missing and why.
- When you have sufficient context, execute the work step by step.

Template 3 — Reusable Manual Prompt Structure: The Brief That Rarely Fails

CONTEXT
- Who I am / what I need / how the result will be used.
- Audience and tone.
OBJECTIVE
- What you want to achieve (in one measurable sentence).
SCOPE
- What IS included and what is NOT (clear boundaries).
DATA / INPUT
- Paste the relevant information here.
- If there are sources, link to them or summarize them.
CONSTRAINTS
- Output format (table, bullets, email, JSON…)
- Approximate length
- Language
- Rules (do not invent, cite sources, etc.)
EXAMPLES
- Example of good output (even if hypothetical).
- Example of bad output (what to avoid).
QUALITY CRITERIA
- 3–7 checks to validate the result.
QUESTIONS
- “Before starting, ask me what you need.”

Model Selection: Which AI for Each Task?

There is no “best model”—only the right one for the problem. In 2026, our technical decision matrix is as follows:

This diversification prevents vendor lock-in and ensures we use the sharpest tool for each cut.

MCPs (Model Context Protocol): the bridge between AI and reality

The biggest qualitative leap at GooApps® has been the implementation of the Model Context Protocol (MCP). Traditional chatbots are isolated; they do not know what is happening inside your company. MCPs act as an open standard that allows AI to securely “read” our internal tools.

1. Atlassian MCP (Jira + Confluence)

Instead of copying and pasting tickets, our AI assistant has read-only access to the backlog.

• Use case: “Analyze ticket JIRA-123, find the business rules in the linked Confluence page, and propose test cases.”
• Result: Full traceability. The AI does not invent requirements; it cites the official source.

2. Figma MCP for development

We connect design context directly to the IDE.

• Use case: The model reads the layer hierarchy and design tokens from Figma to generate CSS/React code that matches pixel-perfectly, reducing frontend “pixel guessing.”

3. Context7 and living documentation

We use MCPs to inject up-to-date technical documentation for fast-changing libraries, preventing the AI from suggesting deprecated methods or non-existent APIs.

Security checklist for AI usage

Connecting AI to internal data requires responsibility. This is our mandatory security checklist:

• Principle of least privilege: the MCP should only have read access to active projects—never to the entire historical repository.

• No PII data: automatic sanitization of personal data before sending any prompt.

• Dependency auditing: strict review of third-party packages used in MCP bridges to prevent supply chain attacks.

• Human-in-the-loop: AI proposes, humans validate. No commit or AI-generated email is sent without human review.

Conclusion: AI as a real competitive advantage

At GooApps®, we do not use AI to “produce more text or code without thinking.” We use it to gain focus, making it work like a good senior professional: asking questions, making assumptions explicit, proposing options, and delivering verifiable results.

By delegating structure, boilerplate, and information retrieval to connected agents (MCPs), our engineers and consultants reclaim time for what truly matters: architecture, product strategy, and the final quality delivered to the client.