MCP - Server Configuration & Integration

SERVER CATEGORIES

Core Development: git, serena, ide | Documentation: context7, readwise, sequential-thinking | Database: supabase | Testing: playwright, puppeteer | Monitoring: sentry, gmail | Deployment: vercel

AGENT-MCP COORDINATION

Utility Agent Delegations (MANDATORY)

• file-creator: File operations (Write, MultiEdit, Edit)
• git-workflow: Version control (git.git_commit, git.git_add, git.git_status)
• knowledge-fetcher: Research (readwise, context7, WebSearch)
• date-checker: Temporal calculations (date commands, filtering)
• context-fetcher: Documentation (Read, Glob, WebFetch)

Engineering Agent Patterns (2024-2025)

Master Template Architecture:

• All Language-Specific Developers: serena (code analysis) + sequential-thinking (complex reasoning) + git (version control)
• Backend Architects: serena + sequential-thinking + git + context7 (architectural patterns)
• Frontend Developers: serena + sequential-thinking + git + playwright (UI testing)

Language-Specific MCP Coordination:

• typescript-node-developer: serena + git + sequential-thinking + context7 (TypeScript/Node.js docs)
• python-backend-developer: serena + git + sequential-thinking + context7 (Python/FastAPI docs)
• nodejs-backend-developer: serena + git + sequential-thinking (JavaScript-specific optimization)
• rust-backend-developer: serena + git + sequential-thinking + context7 (Rust ecosystem docs)
• go-backend-developer: serena + git + sequential-thinking + context7 (Go patterns)

Specialized Problem Solving:

• super-hard-problem-developer: ALL available MCPs (comprehensive toolset for complex debugging)
• refactoring-specialist: serena + git + sequential-thinking (code transformation focus)

Traditional Agent Patterns:

• Testing: test-writer-fixer → playwright/ide → validation
• Code Analysis: Engineering agents → serena/sequential-thinking → insights
• Error Resolution: backend-architect → sentry/supabase → diagnosis
• Deployment: devops-automator → vercel/git → monitoring

AGENT-SPECIFIC MCP ACCESS PATTERNS

Engineering Department MCP Requirements

general_engineering_agents:
  base_requirements: [git, serena, sequential-thinking]
  optional_additions: [context7, playwright]
  
language_specific_developers:
  base_requirements: [git, serena, sequential-thinking]
  documentation_access: [context7]
  restrictions: [NO supabase, NO sentry, NO playwright]
  reasoning: "Focus on code implementation, not infrastructure"
  
specialized_problem_solving:
  super_hard_problem_developer:
    access: "ALL available MCPs"
    reasoning: "Complex problems may require any tool combination"
  refactoring_specialist:
    requirements: [git, serena, sequential-thinking]
    focus: "Code transformation and technical debt reduction"

Design & Marketing Agent Restrictions

design_agents:
  allowed: [git, sequential-thinking, context7, readwise]
  restricted: [serena, supabase, sentry, playwright]
  reasoning: "Visual focus, not code analysis"
  
marketing_agents:
  allowed: [git, sequential-thinking, context7, readwise]
  restricted: [serena, supabase, sentry, playwright]
  reasoning: "Content strategy, not technical implementation"

QUERY CLASSIFICATION RULES

Task Type Detection

simple_lookup:
  indicators: ["find", "get", "show", "list", "what is", "when did"]
  rule: "Use most direct tool, STOP after definitive answer"
  max_tools: 1-2
  
complex_analysis:
  indicators: ["analyze", "compare", "synthesize", "recommend", "explain why", "how should"]
  rule: "Multi-tool coordination acceptable, sequential-thinking encouraged"
  max_tools: 3-5
  
language_specific_development:
  indicators: ["TypeScript backend", "Python API", "Rust service", "Go microservice", "Node.js optimization"]
  agent_routing: "Language-specific developer + base MCP requirements"
  pattern: "serena + git + sequential-thinking + context7"

ERROR RECOVERY & FALLBACKS

• git → Manual git commands → Note version control limitations
• context7 → WebSearch → Manual documentation lookup
• sequential-thinking → Native analysis → Note complexity limitations
• puppeteer/playwright → Manual testing → Provide test cases and fallback instructions
• serena → Text-based code analysis → Note semantic analysis limitations
• sentry → Manual error tracking → Log analysis and issue documentation
• supabase → Manual SQL operations → Database connection alternatives
• vercel → Manual deployment → CI/CD pipeline alternatives
• readwise → Manual search → Note knowledge management gaps
• gmail → Manual email operations → Note communication workflow disruption
• ide → Text-based diagnostics → Note development environment limitations

ANTI-PATTERNS & USAGE WARNINGS

Critical Avoid Conditions

• supabase_operations: read_only_environment, production_lockdown
• browser_automation: headless_server_environment, rate_limited_apis
• sentry_operations: insufficient_permissions, service_outage
• git_operations: detached_head_state
• sequential_thinking: simple_single_step_tasks

Performance Anti-Patterns

• excessive_sequential_thinking_calls: >3 calls per task
• browser_automation_for_api_tasks: Use direct API calls when available
• readwise_export_overuse: STOP after list_highlights unless full content specifically needed
• unnecessary_tool_escalation: Apply STOP rule from query classification

PERFORMANCE OPTIMIZATION

Stop Conditions

stop_execution_when:
  query_answered: "Got definitive answer to user's specific question"
  token_threshold_reached: "simple_lookup: >1000 tokens, medium_analysis: >5000 tokens, complex_analysis: >15000 tokens"
  completion_criteria_met: "Task objectives fully satisfied"
  error_resolution_complete: "Problem identified and solution provided"

Agent-Optimized MCP Performance

Engineering Agent Performance Profiles:

language_specific_developers:
  serena_operations: "100-500ms (code analysis), essential for all implementations"
  git_operations: "50-200ms (version control), required for all code changes"
  sequential_thinking: "1000-5000ms (architectural decisions), for complex problems"
  context7_operations: "300-1000ms (documentation), for framework guidance"
  
specialized_agents:
  super_hard_problem_developer: "All MCPs available, 10-60 second complex analysis cycles"
  refactoring_specialist: "serena + git focus, 2-10 second rapid iteration cycles"

Traditional Tool Performance Profiles:

• git_operations: 50-200ms, minimal tokens, parallel_safe
• supabase_operations: 200-2000ms, low-medium tokens, NOT parallel_safe
• sentry_operations: 300-30000ms, low-very_high tokens, parallel_safe
• browser_automation: 800-5000ms, medium tokens, resource_intensive, NOT parallel_safe
• sequential_thinking: 1000-10000ms, high-very_high tokens, parallel_safe
• serena_operations: 100-2000ms, low-medium tokens, parallel_safe
• readwise_operations: 300-10000ms, low-very_high tokens, parallel_safe

Master Template Efficiency Benefits

consistency_gains:
  all_language_developers: "Same base MCP pattern reduces decision overhead"
  standardized_workflows: "E-H-A-E-D-R cycles optimize MCP usage sequences"
  
performance_optimization:
  focused_tool_sets: "Language developers avoid irrelevant MCPs"
  specialized_expertise: "Deep knowledge reduces trial-and-error MCP usage"
  context_preservation: "Agent spawning eliminates MCP context pollution"