MCP Documentation Skill

⚠️ DEPRECATED: MCP Documentation Skill

STATUS: This skill is deprecated and should not be used.

Why This Skill Was Deprecated

During the MCP cleanup (December 2025), we removed custom documentation caching servers due to:

1. Stale Cache Issues: Cached content became outdated as files changed
2. Complexity Overhead: Custom server maintenance vs. native file access
3. Redundancy: Direct file reading is simpler and more reliable
4. Single Source of Truth: Reading files directly ensures accuracy

Servers That Were Removed

• ❌ CodebaseDocumentation (documentation-server.py) - Moved to _legacy_archive/
• ❌ All custom documentation indexing and caching

Migration Guide

Old Approach (Using This Deprecated Skill)

method: search_docs("deployment")
Risk: Returns cached (potentially stale) content
Token Claim: "90% token savings"
Reality: Cache invalidation issues outweigh benefits

New Approach (Recommended)

# Direct file access (always current)
view_file /path/to/DOCUMENTATION.md

# Search across documentation
grep_search "deployment" in .claude/docs/

# Find specific docs
find_by_name "*.md" in .claude/

# List available agents
list_dir /home/njd/careercopilot/careercopilot-1/.claude/agents/

# List available skills  
list_dir /home/njd/careercopilot/careercopilot-1/.claude/skills/

What To Use Instead

Current Documentation Structure

careercopilot-1/
├── .claude/
│   ├── agents/              # 20 agent definitions
│   ├── skills/              # 29 skill directories
│   ├── docs/                # Architecture docs
│   └── workflows/           # Workflow definitions
├── docs/                    # Project documentation
│   ├── architecture/
│   ├── archive_legacy_reports/
│   └── archive_mcp_configs/
└── README.md                # Main project documentation

Accessing Documentation

List All Agents

// Old: mcp_documentation_get_agents()
// New:
list_dir("/home/njd/careercopilot/careercopilot-1/.claude/agents/")
// Then view specific agents:
view_file("/home/njd/careercopilot/careercopilot-1/.claude/agents/devops-specialist.md")

List All Skills

// Old: mcp_documentation_get_skills()
// New:
list_dir("/home/njd/careercopilot/careercopilot-1/.claude/skills/")
// Then view specific skill:
view_file("/home/njd/careercopilot/careercopilot-1/.claude/skills/component-builder/SKILL.md")

Search Documentation

// Old: mcp_documentation_search_docs("deployment")
// New:
grep_search({
  SearchPath: "/home/njd/careercopilot/careercopilot-1/.claude/docs",
  Query: "deployment",
  Includes: ["*.md"],
  MatchPerLine: true
})

Find Workflows

// Old: Not available in old MCP
// New:
list_dir("/home/njd/careercopilot/careercopilot-1/.agent/workflows/")
view_file("/home/njd/careercopilot/careercopilot-1/.agent/workflows/component_builder.md")

Current MCP Ecosystem

For actual MCP capabilities, use:

1. flash-sidekick MCP - AI assistance

   • consult_pro - Deep reasoning/coding (Pro 2.5)
   • quick_summarize - Fast text summarization (Flash Lite)
   • generate_idf - Python IDF generation (Flash Lite)

2. GitHub MCP - Repository operations

   • File management, PR/issues, search

3. Playwright MCP - Browser automation

   • Via browser_subagent tool

4. Docker MCP - Container management

Performance Reality Check

Token Savings Claim: "90-95% per request"

Reality:

• Native file reading is fast (<100ms)
• No cache invalidation complexity
• Always shows current content
• Simpler mental model
• No custom server maintenance

The "token savings" were theoretical and didn't account for:

• Cache invalidation overhead
• Stale content debugging
• Server maintenance complexity
• CI/CD integration issues

Related Documentation

• See conversation 0b3a6c3f-6c7e-4743-bc7c-a34b3bbe08e3 for MCP cleanup
• Current MCP configuration: mcp.json
• Archived server: _legacy_archive/documentation-server.py

Last Updated: 2025-12-28
Deprecated: 2025-12-27
Replacement: Native file system tools (view_file, grep_search, list_dir, find_by_name)