Discover available tools on-demand. Triggers: what tools, find tool, available MCP, tool for, capability search.
Discover and describe available tools without loading them all into context.
Instead of loading all MCP tools and skills into base context (which burns tokens), this skill discovers them on-demand when you need a specific capability.
# Find all available skills
find .claude/skills -name "SKILL.md" -exec grep -l "" {} \;
# Search skill descriptions for capability
grep -r "description:" .claude/skills/*/SKILL.md | grep -i "<keyword>"
# 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"
# List available agents
ls .claude/agents/*.md 2>/dev/null
# Search agent descriptions
grep -l "<capability>" .claude/agents/*.md
Always available:
Read - Read filesWrite - Create/overwrite filesEdit - Modify filesGlob - Find files by patternGrep - Search file contentsBash - Execute commandsWebSearch - Search the webWebFetch - Fetch URL contentTask - Spawn subagentsLSP - Code intelligence (if configured)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)
This skill implements progressive disclosure:
Token savings: ~85% vs loading all tool schemas upfront
When orchestrator needs to find capabilities:
User: "I need to deploy to Cloud Run"
Search:
grep -ri "cloud run\|gcp\|deploy" .claude/skills/*/SKILL.md .claude/agents/*.md
Result:
cloud-run-deploy at .claude/skills/.../cloud-run-deploy/SKILL.mddevops-engineer for complex deploymentsAction: Load cloud-run-deploy skill, proceed with deployment