Refactor high-complexity React components in Dify frontend. Use when `pnpm analyze-component...
npx skills add TheSimpleApp/agent-skills --skill "mcp-check"
Install specific skill from multi-skill repository
# Description
Verifies MCP server availability and toolkit connections. Run at session start or when tools aren't working. Checks Rube, Firecrawl, browser automation, and other integrations.
# SKILL.md
name: mcp-check
description: Verifies MCP server availability and toolkit connections. Run at session start or when tools aren't working. Checks Rube, Firecrawl, browser automation, and other integrations.
license: MIT
metadata:
author: thesimpleapp
version: "1.0"
MCP Health Check
Verify MCP servers are installed and toolkit connections are active.
When to Run
- Session start - Verify all expected tools are available
- Before complex workflows - Ensure required integrations work
- When tools fail - Diagnose connection issues
- After MCP config changes - Verify new servers loaded
Quick Check
┌─────────────────────────────────────────────────┐
│ MCP HEALTH CHECK │
├─────────────────────────────────────────────────┤
│ │
│ Core Tools: │
│ ├─ Bash, Read, Write, Edit [Built-in ✓] │
│ ├─ Glob, Grep [Built-in ✓] │
│ └─ WebFetch, WebSearch [Built-in ✓] │
│ │
│ MCP Servers: │
│ ├─ Rube/Composio [Check...] │
│ ├─ Firecrawl [Check...] │
│ ├─ Claude in Chrome [Check...] │
│ └─ Marionette (Flutter) [Check...] │
│ │
│ Toolkit Connections (via Rube): │
│ ├─ Fireflies [Check...] │
│ ├─ GitHub [Check...] │
│ ├─ Slack [Check...] │
│ └─ Google (Gmail, Sheets) [Check...] │
│ │
└─────────────────────────────────────────────────┘
Check Procedure
Step 1: Check Rube Availability
TRY:
RUBE_SEARCH_TOOLS
queries: [{use_case: "health check"}]
session: {generate_id: true}
IF SUCCESS:
→ Rube is installed and running
→ Check toolkit_connection_statuses in response
→ Note which toolkits are connected
IF ERROR (tool not found):
→ Rube MCP not installed
→ User needs to add to MCP config
→ Provide setup instructions
Step 2: Check Firecrawl
TRY:
FIRECRAWL_SCRAPE
url: "https://example.com"
formats: ["markdown"]
IF SUCCESS:
→ Firecrawl working
→ Note remaining credits from response
IF ERROR:
→ Firecrawl not configured or out of credits
Step 3: Check Browser Automation
TRY:
mcp__claude-in-chrome__tabs_context_mcp
IF SUCCESS:
→ Browser automation available
→ Note open tabs
IF ERROR:
→ Chrome extension not running
→ User needs to open Chrome with extension
Step 4: Check Marionette (if Flutter project)
TRY:
Check if marionette_mcp available in tool list
IF AVAILABLE:
→ Ready for Flutter app testing
IF NOT:
→ Refer to marionette setup in global rules
Output Format
All Good
✓ MCP Health Check Complete
Core Tools: All available
MCP Servers:
✓ Rube/Composio - Active
✓ Firecrawl - Active (200,418 credits remaining)
✓ Claude in Chrome - Active
✓ Marionette - Available
Toolkit Connections (via Rube):
✓ Fireflies - Connected
✓ GitHub - Connected
○ Fathom - Not connected (optional)
○ Slack - Not connected (optional)
Ready for: meeting-to-tasks, orchestrator, browser automation
Needs Setup
⚠ MCP Health Check - Action Needed
Core Tools: All available
MCP Servers:
✗ Rube/Composio - Not installed
✓ Firecrawl - Active
✓ Claude in Chrome - Active
Action Required:
Rube provides 500+ app integrations (Fireflies, Slack, GitHub, etc.)
To install:
1. Get API key from https://composio.dev
2. Add to your .mcp.json:
"rube": {
"command": "npx",
"args": ["-y", "rube-mcp"],
"env": {
"COMPOSIO_API_KEY": "${COMPOSIO_API_KEY}"
}
}
3. Add COMPOSIO_API_KEY to your .env.local
4. Restart Claude Code
Current capabilities: Firecrawl scraping, browser automation
Missing capabilities: Fireflies API, Slack, GitHub via Rube
Connection Needed
⚠ MCP Health Check - Connection Needed
MCP Servers: All installed ✓
Toolkit Connections:
✗ Fireflies - Not connected
To connect Fireflies:
Run: RUBE_MANAGE_CONNECTIONS toolkit="fireflies"
This will open OAuth flow in browser.
After connecting, meeting-to-tasks will use the API
for richer transcript data.
Rube Setup Instructions
Install Rube MCP
- Get Composio API Key
- Go to https://composio.dev
- Sign up / Log in
-
Copy your API key
-
Add to MCP Config
Windows: %USERPROFILE%\.config\claude-code\mcp.json
Mac/Linux: ~/.config/claude-code/mcp.json
json
{
"mcpServers": {
"rube": {
"command": "npx",
"args": ["-y", "rube-mcp"],
"env": {
"COMPOSIO_API_KEY": "your-key-here"
}
}
}
}
- Restart Claude Code
- Close and reopen terminal
-
Or run
/mcpto reload -
Verify
- Run
/mcp-check - Should show Rube as active
Connect Toolkits
Once Rube is installed, connect individual services:
# Fireflies (meeting transcripts)
RUBE_MANAGE_CONNECTIONS toolkit="fireflies"
# GitHub (repos, issues, PRs)
RUBE_MANAGE_CONNECTIONS toolkit="github"
# Slack (messages, channels)
RUBE_MANAGE_CONNECTIONS toolkit="slack"
# Google Workspace (Gmail, Sheets, Drive)
RUBE_MANAGE_CONNECTIONS toolkit="googlesheets"
RUBE_MANAGE_CONNECTIONS toolkit="gmail"
Each opens OAuth → authorize → stays connected.
Integration
With project-init
/project-init runs:
1. Detect project type
2. /mcp-check → Verify tools for this project type
3. Suggest missing connections
With meeting-to-tasks
/meeting-to-tasks runs:
1. /mcp-check (quick) → Is Fireflies connected?
2. If yes → Use API
3. If no → Fallback to Firecrawl
4. Notify user of upgrade path
With orchestrator
/orchestrator runs:
1. /mcp-check → What tools are available?
2. Plan tasks based on available integrations
3. Skip or fallback for missing tools
Automatic Recovery
When a tool fails mid-workflow:
1. Catch the error
2. Identify which MCP tool failed
3. Check if it's a connection issue vs server issue
4. If connection: Offer to reconnect
5. If server: Suggest fallback approach
6. Continue workflow with available tools
Config in GitHub Repo
The templates/ folder includes MCP config templates:
templates/
├── .mcp.json.template # MCP server config with env vars
└── .env.template # All required environment variables
Copy these to your project and fill in API keys.
The config uses ${ENV_VAR} syntax so secrets stay in .env.local (gitignored).
# Supported AI Coding Agents
This skill is compatible with the SKILL.md standard and works with all major AI coding agents:
Learn more about the SKILL.md standard and how to use these skills with your preferred AI coding agent.