Model Context Protocol - Uncover Root Causes Using Problem Tree Analysis

The Four-Block MCP Structure

Model Context Protocol organizes your AI prompt into four distinct blocks. Each block serves a specific purpose and builds on the previous one:

Block 1: System Context

Define the AI's role and expected behaviors

Block 2: Knowledge Context

Share what you already know about the problem

Block 3: Task Context

Define your research objectives clearly

Block 4: Final Prompt

Clear instruction on what to generate

How MCP Blocks Flow Together

graph TD
    %% INPUT BLOCKS
    B1["📋 BLOCK 1: System Context<br/>Define the AI's Role<br/>• Evidence-focused<br/>• Neutral & transparent<br/>• Cite sources"]
    B2["🧠 BLOCK 2: Knowledge Context<br/>Share What You Know<br/>• Draft problem statement<br/>• Geographic scope<br/>• Existing insights"]
    B3["🎯 BLOCK 3: Task Context<br/>Define Your Objectives<br/>• Problem Tree structure<br/>• Indicator suggestions<br/>• Reading list<br/>• Uncertainties"]
    B4["💬 BLOCK 4: Final Prompt<br/>Clear Instruction<br/>Tie everything together<br/>Request specific outputs"]

    %% PROCESSING
    PROCESS["🤖 AI PROCESSING<br/>Structured Analysis"]

    %% OUTPUT CATEGORIES
    OUT1["📊 Preliminary<br/>Problem Tree"]
    OUT2["📈 Indicator<br/>Suggestions"]
    OUT3["📚 Reading List<br/>with Citations"]
    OUT4["❓ Uncertainties &<br/>Assumptions"]
    OUT5["💡 Stakeholder<br/>Questions"]

    %% Flow relationships
    B1 --> PROCESS
    B2 --> PROCESS
    B3 --> PROCESS
    B4 --> PROCESS

    PROCESS --> OUT1
    PROCESS --> OUT2
    PROCESS --> OUT3
    PROCESS --> OUT4
    PROCESS --> OUT5

    %% Festa Design System Colors
    style B1 fill:#F59E0B,stroke:#D97706,stroke-width:2px,color:#1F2937
    style B2 fill:#F59E0B,stroke:#D97706,stroke-width:2px,color:#1F2937
    style B3 fill:#F59E0B,stroke:#D97706,stroke-width:2px,color:#1F2937
    style B4 fill:#F59E0B,stroke:#D97706,stroke-width:2px,color:#1F2937
    style PROCESS fill:#10B981,stroke:#059669,stroke-width:3px,color:#fff
    style OUT1 fill:#72B043,stroke:#5A8F36,stroke-width:2px,color:#fff
    style OUT2 fill:#72B043,stroke:#5A8F36,stroke-width:2px,color:#fff
    style OUT3 fill:#72B043,stroke:#5A8F36,stroke-width:2px,color:#fff
    style OUT4 fill:#72B043,stroke:#5A8F36,stroke-width:2px,color:#fff
    style OUT5 fill:#72B043,stroke:#5A8F36,stroke-width:2px,color:#fff

The four input blocks feed into AI processing, which generates five structured output categories.

Block 1: System Context (Define the AI's Role)

This block sets the AI's "personality" and establishes quality standards.

Block 1 - System Context

You are an evidence-focused research assistant for nonprofits and social enterprises.

Behaviors: Be neutral, transparent, and concise. Always cite publisher + year for sources.
Prefer reputable sources (UN agencies, government statistics, peer-reviewed research,
established NGOs). Clearly separate facts from assumptions. Flag uncertainties and
potential bias in sources.

Why this matters: Without this framing, AI defaults to general web content. This block ensures outputs prioritize credible development research and maintain transparency about evidence quality.

Block 2: Knowledge Context (Share What You Know)

This block prevents AI from giving you information you already have. It focuses the research on filling gaps.

Block 2 - Knowledge Context

Draft core problem: [Your 1-2 sentence problem statement]
Geographic/population scope: [Location, demographic details if relevant]

What we already know:
• [Existing insight 1 with source if you have it]
• [Existing insight 2]
• [Existing insight 3]

Keywords and related terms: [List of relevant terms, synonyms, related concepts]

Example:

Example - Knowledge Context

Draft core problem: Young adults aged 18-25 in rural Kenya have limited access to
decent employment opportunities.

Geographic/population scope: Rural areas of Nyanza region, Kenya. Youth 18-25 years old,
mixed gender.

What we already know:
• National youth unemployment rate ~20% (Kenya National Bureau of Statistics, 2022)
• Agriculture is declining as primary livelihood
• Some anecdotal evidence of skills-job mismatch

Keywords: youth unemployment, rural livelihoods, skills gap, Kenya Nyanza,
agricultural transition, informal sector

Block 3: Task Context (Define Your Objectives)

This is the most important block—it tells AI exactly what outputs you need.

Block 3 - Task Context

Research objectives:

1) Create a preliminary Problem Tree: core problem statement; 2-3 levels of root causes;
   key effects/consequences

2) Suggest measurable indicators for each major cause and effect (indicator name • unit
   of measurement • typical data sources)

3) Provide 5-10 credible sources for deeper reading (publisher • year • link if available)

4) Identify uncertainties, knowledge gaps, and assumptions that need stakeholder validation

5) Draft 10 targeted questions for stakeholder interviews/focus groups

Why list 5 objectives? This ensures you get everything needed for your Problem Tree in one request: the tree structure, indicators for M&E, sources to cite, gaps to validate, and stakeholder questions.

Block 4: Final Prompt (Clear Instruction)

This block ties everything together with a clear instruction on format.

Block 4 - Final Prompt

Using all the context provided above, generate:

- **Preliminary Problem Tree** in clear markdown format
- **Indicator suggestions** in table format
- **Recommended reading list** with full citations
- **List of uncertainties and assumptions** requiring validation
- **10 stakeholder validation questions** that are open-ended and non-leading

Why specify format? You want outputs you can copy-paste into your documentation. Requesting markdown bullets, tables, and lists makes AI outputs immediately usable.

Complete MCP Template (Copy & Customize)

Here's the full template ready for you to copy, paste into your AI tool, and customize:

Complete MCP Template - Ready to Use

## SYSTEM CONTEXT
You are an evidence-focused research assistant for nonprofits and social enterprises.
Behaviors: neutral, transparent, concise. Cite publisher + year. Prefer reputable sources
(UN agencies, government statistics, multilaterals, peer-reviewed). Separate facts from
assumptions. Flag uncertainties and potential bias.

## KNOWLEDGE CONTEXT
Draft core problem: {{Insert your problem statement}}
Scope/Context: {{Geography, population, timeframe}}
What we already know:
• {{Existing knowledge point 1}}
• {{Existing knowledge point 2}}
• {{Existing knowledge point 3}}
Keywords/aliases: {{Related terms}}

## TASK CONTEXT
Objectives:
1) Propose a Preliminary Problem Tree: core problem; 2-3 levels of root causes; key effects
2) Suggest common indicators that evidence each cause/effect (name • unit • typical sources)
3) Provide 5-10 credible sources to read next (publisher • year • link if available)
4) List uncertainties/gaps and assumptions that need field validation
5) Draft 10 stakeholder research questions to validate and deepen understanding

## PROMPT
Using the contexts above, produce:
- A **Preliminary Problem Tree** in markdown bullets
- A table of **indicators** (Indicator • What it measures • Possible sources)
- A **Reading list** (publisher • year • link if available)
- **Uncertainties & assumptions** (bullets)
- **10 stakeholder questions**

How to Use This Template

1. Copy the entire template above
2. Replace all {{placeholders}} with your specific context
3. Paste into ChatGPT, Claude, or your preferred AI tool
4. Review the output and verify 3-5 key sources
5. Tag findings as (E) evidence or (A) assumption in your Problem Tree

Quality Assurance for AI Outputs

Never accept AI outputs at face value. Run them through this three-checkpoint verification workflow:

graph TD
    %% Start
    START["Receive AI Output<br/>from MCP Prompt"]

    %% CHECKPOINT 1: Source Credibility
    CHECK1{"Source Credibility<br/>Check"}
    Q1A["Are sources reputable?<br/>UN, gov stats,<br/>peer-reviewed, NGOs"]
    Q1B["Are they recent<br/>enough for context?"]
    Q1C["Do links work and<br/>match descriptions?"]
    FAIL1["⚠️ Flag Issues<br/>Request new sources<br/>or verify manually"]

    %% CHECKPOINT 2: Context Fit
    CHECK2{"Context Fit<br/>Assessment"}
    Q2A["Do findings match<br/>your geography/<br/>population?"]
    Q2B["Are cultural & economic<br/>contexts appropriate?"]
    Q2C["Any obvious<br/>contradictions?"]
    FAIL2["⚠️ Flag Mismatches<br/>Adjust MCP prompt<br/>or seek local sources"]

    %% CHECKPOINT 3: Evidence Clarity
    CHECK3{"Evidence vs<br/>Assumption Check"}
    Q3A["Are claims<br/>properly supported?"]
    Q3B["Are limitations<br/>acknowledged?"]
    Q3C["Is bias potential<br/>noted?"]
    FAIL3["⚠️ Tag Assumptions<br/>Mark for stakeholder<br/>validation"]

    %% Success Path
    SUCCESS["✅ Verified Output<br/>Tag evidence (E)<br/>Tag assumptions (A)<br/>Build Problem Tree"]

    %% Relationships
    START --> CHECK1
    CHECK1 -->|Review| Q1A
    Q1A -->|Pass| Q1B
    Q1B -->|Pass| Q1C
    Q1C -->|Pass| CHECK2
    Q1A -->|Fail| FAIL1
    Q1B -->|Fail| FAIL1
    Q1C -->|Fail| FAIL1
    FAIL1 --> START

    CHECK2 -->|Review| Q2A
    Q2A -->|Pass| Q2B
    Q2B -->|Pass| Q2C
    Q2C -->|Pass| CHECK3
    Q2A -->|Fail| FAIL2
    Q2B -->|Fail| FAIL2
    Q2C -->|Fail| FAIL2
    FAIL2 --> START

    CHECK3 -->|Review| Q3A
    Q3A -->|Pass| Q3B
    Q3B -->|Pass| Q3C
    Q3C -->|Pass| SUCCESS
    Q3A -->|Concerns| FAIL3
    Q3B -->|Concerns| FAIL3
    Q3C -->|Concerns| FAIL3
    FAIL3 --> SUCCESS

    %% Festa Design System Colors
    %% Start - Neutral
    style START fill:#6B7280,stroke:#4B5563,stroke-width:2px,color:#fff

    %% Checkpoints - Pot of Gold (decision points)
    style CHECK1 fill:#F59E0B,stroke:#D97706,stroke-width:2px,color:#1F2937
    style CHECK2 fill:#F59E0B,stroke:#D97706,stroke-width:2px,color:#1F2937
    style CHECK3 fill:#F59E0B,stroke:#D97706,stroke-width:2px,color:#1F2937

    %% Questions - Leaf (evaluation criteria)
    style Q1A fill:#72B043,stroke:#5A8F36,stroke-width:1px,color:#fff
    style Q1B fill:#72B043,stroke:#5A8F36,stroke-width:1px,color:#fff
    style Q1C fill:#72B043,stroke:#5A8F36,stroke-width:1px,color:#fff
    style Q2A fill:#72B043,stroke:#5A8F36,stroke-width:1px,color:#fff
    style Q2B fill:#72B043,stroke:#5A8F36,stroke-width:1px,color:#fff
    style Q2C fill:#72B043,stroke:#5A8F36,stroke-width:1px,color:#fff
    style Q3A fill:#72B043,stroke:#5A8F36,stroke-width:1px,color:#fff
    style Q3B fill:#72B043,stroke:#5A8F36,stroke-width:1px,color:#fff
    style Q3C fill:#72B043,stroke:#5A8F36,stroke-width:1px,color:#fff

    %% Fail paths - Chicken Comb (issues to address)
    style FAIL1 fill:#E12729,stroke:#B91C1C,stroke-width:2px,color:#fff
    style FAIL2 fill:#E12729,stroke:#B91C1C,stroke-width:2px,color:#fff
    style FAIL3 fill:#FCD34D,stroke:#F59E0B,stroke-width:2px,color:#1F2937

    %% Success - Pepper Green (validated output)
    style SUCCESS fill:#10B981,stroke:#059669,stroke-width:3px,color:#fff

Key Insight About FAIL3

The yellow "Tag Assumptions" path (FAIL3) is not actually a failure—it's an expected and healthy part of the process. Stakeholder validation is how we turn assumptions into evidence.

Red Flags to Watch For

If you see these warning signs, dig deeper or discard the output:

❌

Generic responses that could apply anywhere

If the Problem Tree could describe any country or population, it's too vague to be useful.

❌

Outdated sources (check publication dates)

For rapidly changing contexts (e.g., post-pandemic employment), sources from 2018 may be obsolete.

❌

Broken or suspicious links

AI sometimes generates plausible-sounding citations that don't exist. Always verify.

❌

Claims without citations

Every major claim should have a source. If AI says "studies show..." but doesn't cite which studies, tag it (A) for assumption.

❌

Overly confident assertions without acknowledging uncertainty

Good research admits knowledge gaps. If AI presents everything as fact, be skeptical.

The 3-Source Rule

For any claim that will be central to your project design, verify it against at least 3 independent sources. If you can't find 3 credible sources supporting it, tag it (A) for assumption and prepare to validate with stakeholders.

Tips for Better MCP Prompts

1. Be Specific About Geography and Population

Weak: "Youth unemployment in Africa"
Strong: "Young adults aged 18-25 in rural Nyanza region, Kenya, with focus on agricultural transition"

2. Include Your Existing Knowledge

Don't start from zero. Share statistics, reports, or insights you already have. This focuses AI on filling gaps rather than repeating basics.

3. Request Specific Output Formats

Ask for "markdown bullets," "tables," or "numbered lists." This makes outputs immediately copy-pasteable into your documentation.

4. Ask for Assumptions and Gaps

Explicitly request "uncertainties requiring validation." This primes AI to flag limitations in published research.

5. Iterate If Needed

If the first output is too general, follow up with: "Focus specifically on [geographic area] and provide more context-specific causes related to [theme]."

Ready to Apply It?

You now have the complete MCP framework and template. Next, you'll learn the step-by-step process for using MCP to build your complete Problem Tree, from research planning through stakeholder preparation.

Model Context Protocol (MCP)