Tool Search

2. Search MCP Servers

# Check configured MCP servers
cat ~/.claude/settings.json 2>/dev/null | grep -A5 "mcpServers" || echo "No global MCP config"

# Check project MCP config
cat .claude/settings.json 2>/dev/null | grep -A5 "mcpServers" || echo "No project MCP config"

3. Search Agents

# List available agents
ls .claude/agents/*.md 2>/dev/null

# Search agent descriptions
grep -l "<capability>" .claude/agents/*.md

4. List Built-in Tools

Always available:

• Read - Read files
• Write - Create/overwrite files
• Edit - Modify files
• Glob - Find files by pattern
• Grep - Search file contents
• Bash - Execute commands
• WebSearch - Search the web
• WebFetch - Fetch URL content
• Task - Spawn subagents
• LSP - Code intelligence (if configured)

Output Format

When searching for a capability, return:

## Available Tools for [Capability]

### Built-in
- **Tool Name**: Brief description

### Skills
- **skill-name**: Description (location: .claude/skills/...)

### MCP Servers
- **server-name**: Description (if configured)

### Agents
- **agent-name**: Description (for complex tasks)

Progressive Disclosure Pattern

This skill implements progressive disclosure:

1. Discovery (this skill): Find what's available
2. Loading: Read full skill/agent file when needed
3. Execution: Use the loaded capability

Token savings: ~85% vs loading all tool schemas upfront

Integration with Orchestrator

When orchestrator needs to find capabilities:

1. Invoke tool-search for discovery
2. Load only matched skills/agents
3. Execute with minimal context

Example

User: "I need to deploy to Cloud Run"

Search:

grep -ri "cloud run\|gcp\|deploy" .claude/skills/*/SKILL.md .claude/agents/*.md

Result:

• Skill: cloud-run-deploy at .claude/skills/.../cloud-run-deploy/SKILL.md
• Agent: devops-engineer for complex deployments

Action: Load cloud-run-deploy skill, proceed with deployment