mcp-integration
Manage MCP (Model Context Protocol) servers - discover, analyze, execute tools. Use for external service integrations, browser automation, documentation lookup, AI image/video generation, database con
SKILL.md.claude/skills/mcp-integration/SKILL.md
GitHubMCP Integration
Overview
MCP (Model Context Protocol) enables Claude to connect with external tools and services through standardized server interfaces.
Key Benefits:
- Connect to external APIs and services
- Browser automation (Playwright)
- Documentation lookup (context7)
- Design file editing (pencil)
- AI multimodal (Gemini for images/videos)
- Database connections
- Custom integrations
Available MCP Servers
Built-in Servers
| Server | Purpose | Example Tools |
|---|---|---|
context7 |
Documentation lookup | resolve-library-id, query-docs |
playwright |
Browser automation | browser_navigate, browser_click, browser_snapshot |
pencil |
Design file editing | batch_get, batch_design, get_screenshot |
Check Available Servers
# List configured MCP servers
claude mcp list
# Or check settings
cat ~/.claude/settings.json | grep -A 20 "mcpServers"
Tool Naming Convention
mcp__{server}__{tool}
Examples:
mcp__context7__resolve-library-id
mcp__context7__query-docs
mcp__playwright__browser_navigate
mcp__playwright__browser_snapshot
mcp__pencil__batch_get
Common Use Cases
1. Documentation Lookup (context7)
When implementing unfamiliar library/framework:
// Step 1: Resolve library ID
mcp__context7__resolve-library-id("react-query")
// Returns: "/tanstack/query"
// Step 2: Query specific topic
mcp__context7__query-docs({
libraryId: "/tanstack/query",
topic: "useQuery hooks"
})
// Returns: Latest documentation
When to use:
- Implementing new library
- Checking latest API changes
- Finding best practices
2. Browser Automation (playwright)
For E2E testing and web automation:
// Navigate to page
mcp__playwright__browser_navigate({ url: "http://localhost:3000" })
// Take snapshot of current state
mcp__playwright__browser_snapshot()
// Click element
mcp__playwright__browser_click({ selector: "button#submit" })
// Fill form
mcp__playwright__browser_fill_form({
selector: "form#login",
values: { email: "test@test.com", password: "password" }
})
// Take screenshot
mcp__playwright__browser_take_screenshot({ path: "screenshot.png" })
// Close browser
mcp__playwright__browser_close()
When to use:
- E2E testing
- Visual verification
- Form automation
- Screenshot capture
3. Design Files (pencil)
For working with .pen design files:
// Get current editor state
mcp__pencil__get_editor_state()
// Read nodes from design
mcp__pencil__batch_get({ patterns: ["**/Button*"] })
// Modify design
mcp__pencil__batch_design({
operations: [
'newButton=I("parent-id", { type: "button", text: "Click me" })'
]
})
// Get screenshot
mcp__pencil__get_screenshot({ nodeId: "some-node-id" })
When to use:
- UI/UX design work
- Component design
- Visual prototyping
Integration in Super Agent
For PM
When assigning tasks that need external services:
## Task 3.5: Implement Payment Integration
**MCP Required:** Yes
**Server:** stripe (if configured)
Instructions:
1. Use context7 to lookup Stripe SDK docs
2. Implement payment flow
3. Use playwright for E2E testing
For Specialists
Before using MCP tools:
# 1. Check if MCP server is available
claude mcp list
# 2. Check tool schema
# Tools are available as mcp__{server}__{tool}
# 3. Execute tool
# Use directly in Claude conversation
For DevOps/QA
Automated testing with playwright:
// Test flow example
async function testLoginFlow() {
// Navigate
await mcp__playwright__browser_navigate({ url: "http://localhost:3000/login" })
// Fill form
await mcp__playwright__browser_fill_form({
selector: "form",
values: { email: "test@test.com", password: "password" }
})
// Click submit
await mcp__playwright__browser_click({ selector: "button[type=submit]" })
// Verify redirect
const snapshot = await mcp__playwright__browser_snapshot()
// Check snapshot for dashboard content
}
Configuration
Adding MCP Servers
Edit ~/.claude/settings.json:
{
"mcpServers": {
"context7": {
"command": "npx",
"args": ["-y", "@anthropic/context7-mcp"]
},
"playwright": {
"command": "npx",
"args": ["@anthropic/playwright-mcp"]
},
"custom-server": {
"command": "node",
"args": ["/path/to/your/mcp-server.js"],
"env": {
"API_KEY": "your-api-key"
}
}
}
}
Project-Specific Configuration
Add to .claude/settings.json in project root:
{
"mcpServers": {
"project-db": {
"command": "npx",
"args": ["@prisma/mcp-server"],
"env": {
"DATABASE_URL": "${DATABASE_URL}"
}
}
}
}
Error Handling
| Error | Cause | Solution |
|---|---|---|
Server not found |
MCP server not configured | Add to settings.json |
Tool not found |
Wrong tool name | Check available tools |
Connection failed |
Server crashed | Restart Claude Code |
Permission denied |
Missing env vars | Set required env variables |
Best Practices
- Check availability first - Verify MCP server is configured
- Handle errors gracefully - Services may be unavailable
- Use appropriate server - Match tool to task
- Document MCP usage - Note in task logs for knowledge base
- Don't expose secrets - Use env vars for API keys
Quick Reference
| Task | Server | Tool |
|---|---|---|
| Lookup docs | context7 | query-docs |
| Navigate browser | playwright | browser_navigate |
| Click element | playwright | browser_click |
| Take screenshot | playwright | browser_take_screenshot |
| Fill form | playwright | browser_fill_form |
| Get browser state | playwright | browser_snapshot |
| Read design | pencil | batch_get |
| Edit design | pencil | batch_design |
Integration with Other Skills
- Use with
test-driven-developmentfor E2E test implementation - Use with
systematic-debuggingfor browser-based debugging - Use with
visual-previewfor design documentation - Use with
qa-testingfor automated test suites