MCP GoGrab AI — Internal Research & Browsing Protocol

Purpose

This document defines the mandatory process for conducting structured research inside mcp-gograb-ai.

The objective is not simple retrieval.
The objective is high-confidence insight (≥ 0.9 calibrated confidence) grounded in decisions, outcomes, patterns, and system intelligence.

This protocol must be followed for:

• Architectural investigations
• Bug pattern analysis
• Performance regressions
• Decision archaeology
• Agent evolution analysis
• Codebase structural review
• Strategic technical planning

1. Core Principles

1. Research must be structured, not reactive.
2. Every conclusion must be calibrated.
3. Decisions must be correlated with outcomes.
4. Patterns must be validated, not assumed.
5. If confidence < 0.9, continue researching.

2. Phase 1 — Context Initialization (Mandatory)

Always begin with:

preflight({ task: "<clear research objective>" })

If the topic relates to a past issue or theme:

find_similar({ query: "<problem or concept>" })

If prior reasoning or decision memory may exist:

intelligence({ action: "recall", situation: "<context>" })

If session-level history is relevant:

session({ action: "summary" })

Do not skip initialization.

3. Phase 2 — Structured Exploration

Select the appropriate exploration pathway.

A. Decision Archaeology

Use when investigating historical behavior, regressions, pivots, or repeated mistakes.

intelligence({ action: "trace", sessionNumber?: X })
intelligence({ action: "pivots" })
intelligence({ action: "loops" })

Goal:

• Identify pivotal decisions
• Detect repeated anti-patterns
• Map causal chains

B. Lessons & Reasoning Extraction

Use when identifying design principles or past learning.

intelligence({ action: "lessons", topic?: "..." })
intelligence({ action: "explain", event: "...", context: "..." })

Goal:

• Extract principles
• Identify systemic reasoning patterns
• Avoid repeating flawed assumptions

C. Outcome Correlation

Research is incomplete without outcome validation.

stats({ action: "enhanced" })
outcome({ action: "report", decisionId: "..." })
outcome({ action: "module", moduleName: "..." })

Goal:

• Measure decision success rate
• Correlate architecture with performance
• Detect fragile systems

D. Codebase Structural Intelligence

Use when analyzing drift, violations, or architecture integrity.

intelligence({ action: "analyze_dna", projectPath: "..." })
intelligence({ action: "violations" })

Goal:

• Detect structural inconsistencies
• Identify architectural drift
• Validate codebase health

E. Predictive Research

Use before major refactors or architectural changes.

intelligence({
  action: "simulate",
  simulateAction: "...",
  context: "...",
  alternatives: ["..."]
})

If assigning responsibility:

intelligence({
  action: "predict_best",
  taskType: "...",
  taskDescription: "..."
})

Goal:

• Reduce implementation risk
• Forecast trade-offs
• Compare alternatives before acting

4. Phase 3 — Confidence Calibration (Required)

Before presenting conclusions:

intelligence({
  action: "calibrate",
  rawConfidence: 0.X,
  domain: "<angular | architecture | platform | etc>"
})

Interpretation:

• ≥ 0.9 → High confidence, can recommend action
• 0.75–0.89 → Likely, requires validation
• < 0.75 → Insufficient evidence, continue research

No major recommendation should be made without calibrated confidence.

5. Phase 4 — Insight Synthesis (Mandatory Output Structure)

Every research result must include:

1. Validated Facts

Concrete, traceable information from MCP.

2. Identified Patterns

Repeated themes, architectural tendencies, anti-patterns.

3. Risk Assessment

• Fragile areas
• Blind spots
• Drift from conventions
• Repeated failure loops

4. Reinforced Decisions

Which past decisions are confirmed as correct?

5. Decisions That Should Be Reconsidered

Which decisions show weak outcomes or recurring issues?

6. Calibrated Confidence Score

Explicit value after calibration.

Research is incomplete without structured synthesis.

6. Anti-Patterns (Do Not Do This)

• Do not stop after a single find_similar call.
• Do not present intuition as evidence.
• Do not skip outcome correlation.
• Do not skip calibration.
• Do not ignore loops or pivots.
• Do not assume success without stats validation.

7. When to Escalate

Escalate research depth if:

• A pattern appears more than twice.
• A decision impacts multiple modules.
• An issue reoccurs across sessions.
• Confidence remains < 0.9 after initial exploration.

Escalation tools:

intelligence({ action: "reflect", taskSummary: "...", outcome: "...", lessonsLearned: [...] })
intelligence({ action: "gaps", domain: "..." })

8. System-Level Goal

Using this protocol should:

• Reduce repeated architectural mistakes
• Increase calibrated decision accuracy
• Detect fragility earlier
• Strengthen reinforcement learning loops
• Increase measurable system intelligence over time

This is not browsing. This is institutional cognition.

9. Minimal Required Flow (Quick Reference)

preflight({ task: "..." })
find_similar({ query: "..." })
intelligence({ action: "recall", situation: "..." })
... structured exploration ...
stats({ action: "enhanced" })
intelligence({ action: "calibrate", rawConfidence: X, domain: "..." })

If confidence < 0.9 → Continue.

If confidence ≥ 0.9 → Conclude with synthesis.

End of Protocol.