Repomix Codebase Explorer

Pack and analyze codebases into AI-optimized formats. Three access methods: MCP tools (maximum control), slash commands (quick access), agents (autonomous exploration).

---

Quick Decision Tree

What do you need?
    │
    ├─ Quick exploration of a repo?
    │       └─ /repomix-explorer:explore-remote (slash command)
    │
    ├─ Precise control over packing/filtering?
    │       └─ MCP tools (pack_codebase, pack_remote_repository)
    │
    ├─ Autonomous multi-step analysis?
    │       └─ repomix-explorer:explorer agent
    │
    ├─ Search within packed output?
    │       └─ MCP grep_repomix_output (after packing)
    │
    └─ Create reusable reference skill?
            └─ MCP generate_skill

---

Access Methods

1. MCP Tools (Maximum Control)

Always check schema first: mcp-cli info repomix/<tool> before calling.

Task	Tool	Example
Pack local directory	pack_codebase	mcp-cli call repomix/pack_codebase '{"directory": "/path/to/project"}'
Pack GitHub repo	pack_remote_repository	mcp-cli call repomix/pack_remote_repository '{"remote": "user/repo"}'
Generate skill from code	generate_skill	mcp-cli call repomix/generate_skill '{"directory": "/path", "skillName": "my-skill"}'
Search packed output	grep_repomix_output	mcp-cli call repomix/grep_repomix_output '{"outputId": "xxx", "pattern": "function.*auth"}'
Read packed output	read_repomix_output	mcp-cli call repomix/read_repomix_output '{"outputId": "xxx", "startLine": 1, "endLine": 100}'
Attach existing pack	attach_packed_output	mcp-cli call repomix/attach_packed_output '{"path": "/path/to/output.xml"}'

See: references/mcp-tools.md for complete schemas.

2. Slash Commands (Quick Access)

Command	Purpose	When to Use
/repomix-explorer:explore-local	Interactive local exploration	Understand local project structure
/repomix-explorer:explore-remote	Interactive GitHub exploration	Quick remote repo analysis
/repomix-commands:pack-local	Pack local codebase	Need packed output without exploration
/repomix-commands:pack-remote	Pack remote repository	Pack GitHub repo quickly

See: references/slash-commands.md for details.

3. Agent Delegation (Autonomous)

// Use Task tool with repomix-explorer:explorer
Task({
  subagent_type: "repomix-explorer:explorer",
  prompt: "Analyze the authentication patterns in https://github.com/user/repo",
  description: "Analyze auth patterns in repo"
})

Best for:

• Complex multi-step analysis requiring autonomous decisions
• When you need a summary rather than raw output
• Parallel analysis across multiple repositories

See: references/agent-integration.md for orchestration patterns.

---

Critical Rules

ALWAYS

Rule	Why
Run mcp-cli info repomix/<tool> before calling	Schemas have required params you might miss
Use grep_repomix_output for targeted search	Don't load entire codebase when you need specific patterns
Store outputId after packing	Required for subsequent grep/read operations
Use includePatterns for large repos	Focus on relevant files, reduce token cost

NEVER

Anti-Pattern	Do Instead
Pack entire repo then read all	Pack once, grep multiple times
Ignore compress for 2000+ file repos	Enable compression OR use incremental grep
Call grep without checking outputId exists	Pack first, store the ID, then grep
Skip pattern filtering on monorepos	Use includePatterns to focus scope

---

Token Optimization

Codebase Size	Strategy
<100 files	Full pack, no compression
100-500 files	Full pack with XML format
500-2000 files	Use compression OR incremental grep
>2000 files	Always incremental grep workflow

Compression reduces ~70% tokens while preserving semantics via Tree-sitter.

---

Integration Points

With cognee-memory

# After analysis, persist findings for future sessions
mcp-cli call cognee/save_interaction '{
  "interaction_type": "analysis",
  "content": "Analyzed auth patterns in user/repo: JWT middleware in /src/auth, refresh tokens in /src/tokens"
}'

With database-operations

Combine schema context with codebase analysis for full-stack understanding:

1. Pack the codebase
2. Query database schema via postgres MCP
3. Cross-reference code patterns with DB structure

With ai-opus-debate

Route architectural decisions discovered during analysis:

# After finding multiple auth patterns
Task({
  subagent_type: "ai-opus-debate",
  prompt: "Debate: JWT vs session-based auth based on patterns found in codebase analysis..."
})

---

Common Workflows

1. Explore Remote Repository

# Quick: Use slash command
/repomix-explorer:explore-remote

# Or MCP for control:
mcp-cli call repomix/pack_remote_repository '{"remote": "user/repo"}'
# Store outputId from response
mcp-cli call repomix/grep_repomix_output '{"outputId": "ID", "pattern": "export.*function"}'

2. Generate Reference Skill

mcp-cli call repomix/generate_skill '{
  "directory": "/path/to/library",
  "skillName": "library-reference",
  "includePatterns": "src/**/*.ts"
}'
# Creates .claude/skills/library-reference/ with SKILL.md and references/

3. Code Review Pipeline

# 1. Pack the PR branch changes
mcp-cli call repomix/pack_codebase '{"directory": "/path", "includePatterns": "src/**"}'

# 2. Search for patterns to review
mcp-cli call repomix/grep_repomix_output '{"outputId": "ID", "pattern": "TODO|FIXME|console\\.log"}'

# 3. Route to Opus for deep review
Task({ subagent_type: "ai-opus-debate", prompt: "Review these patterns..." })

---

Anti-Patterns

Mistake	Consequence	Fix
Packing without includePatterns on monorepo	100k+ token output	Filter to relevant paths
Using read_repomix_output for entire file	Token explosion	Use grep for targeted search
Re-packing same repo multiple times	Wasted API calls	Store and reuse outputId
Ignoring security scan results	Committing secrets	Review security warnings in pack output

---

References

• MCP Tools Reference - Complete parameter schemas
• Slash Commands - Quick command reference
• Agent Integration - Orchestration patterns
• Workflow Guide - Comprehensive tutorials