12. Context Packing Tools

When working with LLMs on large codebases, context packing refers to techniques for extracting and feeding relevant code context to the model efficiently.

Why Context Matters

Claude Code automatically reads files as needed, but external tools exist for:

• Pre-session preparation: Dump relevant code before starting
• Cross-tool workflows: Feed context to models outside Claude Code
• Offline analysis: Prepare context for later use

Available Tools

Tool	Purpose	How It Works
gitingest	Repo → text dump	Extracts relevant files into a single text file for LLM consumption
repo2txt	Repo → formatted context	Similar to gitingest, with formatting options
Context7 MCP	Docs lookup	Fetches library documentation on-demand (see MCP section)

When to Use (and When Not)

Scenario	Best Approach
Working in Claude Code	Let Claude read files naturally (no pre-dumping needed)
Feeding context to ChatGPT/Gemini	gitingest/repo2txt useful
Preparing spec for team review	Export relevant files to share
Very large monorepo (>1M LOC)	May help with selective extraction

Note: Claude Code's native file access is usually sufficient. These tools are most useful for cross-tool workflows or when working with models that don't have file system access.

Source

• Addy Osmani: My AI Coding Workflow in 2026, discussing context packing as part of a broader AI development workflow

Architecture Diagrams as Context (Advanced Pattern)

For large OOP codebases, research confirms LLMs struggle with polymorphism and dependency reasoning when processing files in chunks (ACM 2024: "LLMs Still Can't Avoid Instanceof").

Problem: File chunking loses structural relationships (class hierarchies, interface implementations, cross-module dependencies).

Solution: Include architecture diagrams in project context to provide explicit relationships.

Approaches

Approach	Maintenance	Token Cost	Best For
Archy MCP	Zero (auto-gen)	On-demand	GitHub repos with class hierarchies
Inline Mermaid	Manual	200-500 tokens	Custom architectural views
PlantUML ref	Manual	Minimal	Enterprise/IDE integration

MCP Tools for Architecture Visualization

Archy MCP (phxdev1, April 2025):

• Auto-generates Mermaid from GitHub repos or text descriptions
• Supports: flowcharts, class diagrams, sequence diagrams
• URL: pulsemcp.com/servers/phxdev1-archy

Mermaid MCP (hustcc, 61.4K users):

• Custom themes, background colors
• Real-time rendering

Blueprint MCP (ArcadeAI):

• Text descriptions → technical diagrams
• Async job management

Inline Example (CLAUDE.md)

## Architecture Overview

\`\`\`mermaid
classDiagram
    class UserService {
        +authenticate()
        +getProfile()
    }
    class AuthProvider {
        <<interface>>
        +validate()
    }
    UserService --> AuthProvider
\`\`\`

When to Use

• OOP codebases >20 modules with complex inheritance
• Java/Spring projects with deep polymorphism
• When Serena symbol overview is insufficient

Recommended Workflow

1. Try Serena first: get_symbols_overview + find_symbol (zero maintenance)
2. If insufficient: Use Archy MCP to auto-generate class diagrams
3. Last resort: Manual inline Mermaid for custom views

Key Insight

"Context structure matters more than context size." Explicit relationships improve LLM reasoning on OOP architectures.

Source: LinkedIn discussion (Jan 2026)

Note: Pattern reported on Java/Spring project. Not validated at scale. Alternative Serena + grepai achieves similar results with zero maintenance.