Connect AI: Adding external Systems to your AI Agent / Copilot

Whenever you want to connect external Systems to Connect AI, you have 3 different options.
  1. MCP action
  1. API action (soon for Copilot)
  1. Website search action (soon for AI Agent and Copilot)
💡
We recommend considering the options from top to bottom, meaning that MCP actions should be looked into first, should that not be possible, try API actions, if that also does not work, use Website search.
MCP and API actions require some technical understanding of setting up servers or interfaces, but provide fast and reliable answers. Website search actions on the other hand only need a publicly available website, with the downside of slower response times and risk of breaking due to changes on the website. You can finde more general information with pros and cons on all 3 options below.
⚠️
Disclaimer! The quality of your data really matters! If you provide the AI with low-quality information, you'll get low-quality results back (think "garbage in, garbage out")

MCP actions

Pros

  • Structured protocol: MCP provides a standardized way to connect tools and data sources with consistent interfaces.
  • Rich context: Can access and manipulate structured data, resources, and tools with full context awareness.
  • Bidirectional communication: Supports real-time interactions and state management between client and server.
  • Type safety: Built-in schema validation and type definitions reduce errors.
  • Extensibility: Easy to add new capabilities through prompts, resources, and tools.

Cons

  • Setup complexity: Requires implementing or integrating MCP server infrastructure.
  • Limited adoption: Still a relatively new protocol with fewer existing integrations.
  • Development overhead: Need to build or maintain MCP-compliant servers.
  • Learning curve: Teams need to understand MCP concepts and architecture.
💡
Find our MCP tutorial here

API actions

Pros

  • Simplicity: Easy to implement with standard HTTP requests and responses.
  • Widespread adoption: Almost all services provide REST or similar APIs.
  • Flexible: Can be customized to specific use cases and endpoints.
  • Well-documented: Extensive resources and libraries available for API integration.
  • Direct control: Full control over request parameters and response handling.

Cons

  • Limited structure: Requires manual parsing and interpretation of responses.
  • Inconsistent formats: Different APIs use different conventions and data structures.
  • Authentication complexity: Need to manage API keys, tokens, and authentication flows.
  • Rate limiting: Often subject to usage quotas and throttling.
  • Maintenance burden: API changes require code updates.
💡
Find our API actions tutorial here

Website search actions

Pros

  • Universal access: Can extract data from any publicly accessible website.
  • No API needed: Works even when official APIs don't exist or are restricted.
  • Free data access: No API subscription costs or usage fees.
  • Comprehensive coverage: Access to all visible content on a webpage.

Cons

  • Fragility: Breaks easily when website structure changes.
  • Legal concerns: May violate terms of service or copyright restrictions.
  • Performance issues: Slower than direct API calls, requires full page loading.
  • Data quality: Unstructured HTML requires extensive parsing and cleaning.
  • Bot detection: Websites may block or limit scraping attempts.
  • Maintenance intensive: Selectors and parsing logic need frequent updates.
💡
COMING SOON

Comparison Table

Aspect
MCP Server Connections
Simple API Request
Website Scraping/Search
Ease of Setup
Complex - requires MCP infrastructure
Simple - standard HTTP requests
Moderate - requires parsing logic
Data Structure
Highly structured with schema validation
Structured but varies by API
Unstructured - requires extensive parsing
Adoption
Limited - new protocol
Widespread - industry standard
Universal - works on any website
Communication
Bidirectional with state management
Request-response model
One-way data extraction
Maintenance
Moderate - protocol updates needed
High - API changes require updates
Very high - frequent selector updates
Performance
Fast with real-time capabilities
Fast - direct endpoint access
Slow - full page loading required
Cost
Development and infrastructure costs
API subscription fees and quotas
Free but resource-intensive
Reliability
High - structured protocol
High - official API support
Low - breaks with website changes
Legal Considerations
Clear terms through protocol
Governed by API terms of service
May violate ToS or copyright
Best Use Case
Complex integrations with multiple tools
Standard data access from services
When no API exists