MCP Integration

MCP Integration Skill

Purpose: Guide the integration and effective use of Model Context Protocol (MCP) servers to extend AI agent capabilities with external tools, data sources, and services.

Overview

The Model Context Protocol (MCP) is an open standard that connects AI systems with external tools and data sources through a unified interface. MCP servers act as capability providers — each server exposes a set of tools that the AI can invoke to perform actions beyond its native abilities.

This skill ensures that MCP integrations follow Trust-Grade principles: secure, reliable, and properly governed.

Core Concepts

MCP Architecture

┌─────────────────┐     ┌──────────────────┐     ┌─────────────────┐
│    AI Agent      │────▶│   MCP Client     │────▶│   MCP Server    │
│  (LLM + Kit)    │◀────│  (IDE/Runtime)   │◀────│  (Tool Provider)│
└─────────────────┘     └──────────────────┘     └─────────────────┘
                                                         │
                                                    ┌────┴────┐
                                                    │ External │
                                                    │ Service  │
                                                    └─────────┘

Key Components

Integration Patterns

Pattern 1: Git Operations via MCP

When to use: Git operations that benefit from structured API access rather than raw CLI commands.

Available servers: GitKraken MCP, GitHub MCP

Preferred for:
- Creating branches, commits, PRs (structured data)
- Reading git status, blame, diff (parsed output)
- Managing issues and reviews (API access)

Still use CLI for:
- Complex git operations (rebase, cherry-pick)
- Local-only operations (stash, worktree)
- Performance-critical batch operations

Pattern 2: Issue and Project Management

When to use: Creating, updating, and querying issues across platforms.

GitHub MCP Server provides:
- issue_write (create/update issues)
- issue_read (get details, comments, labels)
- search_issues (query with GitHub syntax)
- list_issues (paginated browsing)

Best practices:
- Use search_* for targeted queries with criteria
- Use list_* for broad retrieval with pagination
- Batch reads in groups of 5-10 items

Pattern 3: Code Search and Analysis

When to use: Finding code patterns across repositories.

GitHub MCP provides:
- search_code (cross-repo code search)
- get_file_contents (read specific files)
- get_commit (commit details with diff)

Usage guidance:
- Use search_code for finding symbols, functions, patterns
- Use get_file_contents for reading specific known files
- Combine with local grep_search for current workspace

Pattern 4: Pull Request Workflow

When to use: Full PR lifecycle management.

Workflow:
1. create_branch → Create feature branch
2. push_files → Push code changes
3. create_pull_request → Open PR
4. request_copilot_review → Automated review
5. pull_request_read(get_status) → Check CI status
6. merge_pull_request → Merge when ready

Trust-Grade rules:
- Always create PR (never push directly to main)
- Always request review before merge
- Always check CI status before merge

Security Guidelines

Authentication

• MCP servers use token-based authentication
• Tokens are managed by the IDE/runtime, never by the AI agent
• Never log, display, or store authentication tokens
• Verify server identity before sending sensitive operations

Permission Model

Data Handling

• Never pass secrets through MCP tool parameters
• Sanitize user data before including in tool calls
• Respect repository visibility (public vs private)
• Log operations for audit trail (session-context.md)

Server Selection Guide

Error Handling

Common MCP Errors

Fallback Strategy

When an MCP server is unavailable:

1. Log the failure in session context
2. Fall back to CLI equivalent if available
3. Notify user of degraded capability
4. Continue with remaining available tools

Integration with Antigravity AI Kit

Loading Rules Integration

MCP capabilities are registered in engine/loading-rules.json under domain rules:

{
  "domain": "git-operations",
  "keywords": ["git", "commit", "branch", "pr", "merge"],
  "loadAgents": ["devops-engineer"],
  "loadSkills": ["git-workflow", "mcp-integration"],
  "mcpServers": ["GitKraken", "github-mcp-server"]
}

Session Context Tracking

Log MCP operations in session context for continuity:

## MCP Operations This Session
- Created branch `feature/auth` via GitKraken MCP
- Opened PR #42 via GitHub MCP
- Requested Copilot review via GitHub MCP

Workflow State Integration

MCP operations map to workflow phases: