Skip to content

Bootstrap Process Example

This document shows a complete example of the automatic bootstrap process when a new repository is set up with the MCP Local RAG system.

User Action

User copies bootstrap package to repository root, then types:

Please bootstrap from MCP_BOOTSTRAP_PACKAGE.md

Phase 1: Root Context Auto-Population

Claude's Action: Automatically detects repository structure and populates the root context.

Discovery Process:

# Claude automatically runs these commands
ls -la                                      # Find top-level directories
find mcp-docs -name "*.pdf" | head -5      # Check for documentation
cat README.md                              # Read project description
cat build.gradle                           # Detect Java/Gradle
grep -r "@SpringBootApplication"           # Find Spring Boot

Claude's Output:

I'm bootstrapping the context system for your repository.

Phase 1: Creating root context...

Auto-detected:
- Repository: payment-processor
- Purpose: (from README.md) "Payment processing service for handling transactions"
- Main modules: api/, core/, processors/, shared/
- Technologies: Java 17, Gradle, Spring Boot, PostgreSQL, Kafka
- Documentation: Found 3 PDFs in mcp-docs/

Writing CLAUDE_CONTEXT.md...
Root context created (152 lines)

Generated File: CLAUDE_CONTEXT.md

# Payment Processor - Claude Context

## MCP Documentation System

> **Full MCP Rules**: See `mcp-docs/claude-code/MCP_STANDARD_RULES.md` for complete documentation access workflow
>
> This repository uses MCP Local RAG for semantic search over payment specifications.

### Repository-Specific MCP Configuration

- **`{{DOC_TYPE}}`**: Payment specifications
- **`{{DOCS_DIR}}`**: `mcp-docs/`
- **`{{SETUP_SCRIPT}}`**: `mcp-docs/claude-code/setup.sh`

**Available Documentation**:
- Payment specs: `mcp-docs/` (3 PDF files found)

---

## Purpose

Payment processing service for handling transactions with support for multiple payment networks.

## Repository Structure

### Main Modules
- **api/**: REST API endpoints for payment requests
- **core/**: Core payment processing logic
- **processors/**: Network-specific payment processors
- **shared/**: Shared utilities and models

[... rest of auto-generated content ...]

Phase 2: Module Context Auto-Creation

Claude's Action: Automatically identifies major modules and creates contexts for each.

Claude's Output:

Phase 2: Creating module contexts...

Identified modules:
1. api/ (REST API layer)
2. core/ (Payment processing core)
3. processors/ (Network processors)
4. shared/ (Shared utilities)

Creating contexts for top 3 modules...

Module 1: api/

# Claude runs discovery commands
cd api
find . -type d -maxdepth 2                 # Package structure
find . -name "*.java" | head -10           # Find key classes
grep -r "@RestController"                  # Find controllers
cat src/main/resources/application.yml     # Check config

Output:

api/CLAUDE_CONTEXT.md created (87 lines)
  - Detected: Spring REST controllers, 3 main endpoints
  - Key classes: PaymentController, AuthController, WebhookController
  - Configuration: Port 8080, security config detected

Module 2: core/

cd core
find . -name "*Service.java"               # Service layer
find . -name "*Repository.java"            # Data layer
grep -r "public class.*Processor"          # Core processors

Output:

core/CLAUDE_CONTEXT.md created (124 lines)
  - Detected: Service layer with 5 services
  - Processing flow: Request → Validation → Authorization → Settlement
  - Data models: Payment, Transaction, AuthResult

Module 3: processors/

cd processors
ls -la                                     # Submodules
find . -name "*.java" | wc -l             # Class count
grep -r "extends.*Processor"              # Processor hierarchy

Output:

processors/CLAUDE_CONTEXT.md created (156 lines)
  - Detected: Network-specific processors
  - Processing flow: Route → Format → Submit → Parse response
  - External documentation: Links to payment specs in mcp-docs/

Claude Asks:

I've created contexts for api/, core/, and processors/ (3 modules).

Should I continue with shared/ module, or are these three sufficient for now?
You can always add more module contexts later using the template.

User responds: "These three are sufficient"

Phase 3: Completion Report

Claude's Output:

Context System Bootstrap Complete

Root Context: CLAUDE_CONTEXT.md (152 lines)

Module Contexts Created:
  - api/CLAUDE_CONTEXT.md (87 lines)
  - core/CLAUDE_CONTEXT.md (124 lines)
  - processors/CLAUDE_CONTEXT.md (156 lines)

MCP Documentation System: Configured
  - Found 3 PDFs in mcp-docs/
  - MCP setup script available: mcp-docs/claude-code/setup.sh
  - Run setup and restart Claude Code to enable documentation queries

Git Status: All files staged for commit
  - mcp-docs/claude-code/ (7 files)
  - CLAUDE_CONTEXT.md
  - api/CLAUDE_CONTEXT.md
  - core/CLAUDE_CONTEXT.md
  - processors/CLAUDE_CONTEXT.md

Next Steps:
1. Review generated contexts for accuracy
2. Run MCP setup: cd mcp-docs && ./claude-code/setup.sh
3. Restart Claude Code
4. Review staged changes: git status
5. Commit when ready: git commit -m "Add Claude Code context system"

Total time: ~3 minutes

What Just Happened?

  1. User ran one command: "Please bootstrap from MCP_BOOTSTRAP_PACKAGE.md"

  2. Claude automatically:

    • Analyzed repository structure
    • Detected technologies and build tools
    • Read README for purpose
    • Found documentation in mcp-docs/
    • Created root CLAUDE_CONTEXT.md with all details
    • Identified 4 major modules
    • Created 3 module-specific contexts (asked about 4th)
    • Staged all created files in git
    • Generated actionable next steps

  3. Result: Complete context system in ~3 minutes with minimal user interaction

  4. Benefits:

    • No manual template filling
    • No guessing about structure
    • Documentation automatically linked
    • Cross-module references included
    • Files automatically staged in git
    • Ready to use immediately

After Bootstrap

User runs MCP setup:

cd mcp-docs
./claude-code/setup.sh

Setup script:

  1. Installs Node.js (if needed)
  2. Installs mcp-local-rag package
  3. Configures ~/.claude.json
  4. Verifies PDF files
  5. Provides ingestion instructions

User restarts Claude Code

MCP Local RAG is now active. Documentation queries work automatically:

User: "What are the field definitions for payment authorization messages?"

Claude:
1. Checks if MCP is configured
2. Queries documentation: mcp__local-rag__query_documents("authorization message fields")
3. Returns results from payment specs

Key Takeaways

Efficiency:

  • Bootstrap: 3 minutes (vs 30+ minutes manual)
  • One user command (vs 10+ manual steps)
  • 3-5 contexts created automatically

Accuracy:

  • Auto-detection from actual code structure
  • No placeholder mistakes
  • Consistent formatting

Maintainability:

  • Templates stay generic
  • Easy to propagate updates
  • Self-documenting system

User Experience:

  • Minimal interaction required
  • Clear progress updates
  • Actionable next steps