MCP Servers
Model Context Protocol (MCP) servers extend AI capabilities by providing custom tools. Build tools that AI assistants can use.
What is MCP?
MCP (Model Context Protocol) is a standard for:
- Exposing tools to AI models
- Providing context and data
- Enabling AI-powered actions
Use Cases
- Connect AI to your APIs
- Create custom AI tools
- Build AI-powered widgets
- Extend AI assistants
How MCP Works
Architecture
AI Model ←→ MCP Server ←→ Your API/Service
- AI model requests a tool
- MCP server handles the request
- Executes against your service
- Returns results to AI
Tools
Tools are functions the AI can call:
- Defined with name and description
- Take structured parameters
- Return structured results
Example tool:
{
name: "get_weather",
description: "Get current weather for a location",
parameters: {
location: { type: "string", required: true }
}
}
Using MCP Servers
Installing from Marketplace
- Go to Marketplace > MCPs
- Find an MCP server
- Click Install
- Configure credentials
Connecting to Workflows
MCP tools can be used in workflows:
- Add an MCP Action node
- Select the installed MCP
- Choose the tool
- Configure parameters
Using with AI Chat
MCPs enhance AI assistants:
- Install MCP server
- Enable for chat
- AI automatically uses available tools
Built-in MCPs
Database MCP
Query your databases via AI:
- Natural language queries
- Automatic SQL generation
- Safe read-only by default
API MCP
Expose your APIs to AI:
- Import from OpenAPI
- AI can call endpoints
- Results formatted for AI
File MCP
File operations:
- Read files
- Search content
- Generate reports
Managing MCPs
View Installed
- Go to Settings > MCPs
- See all installed servers
- View status and usage
Configuration
Each MCP may need:
- API credentials
- Connection settings
- Permission scopes
Permissions
Control what MCPs can do:
- Read-only vs read-write
- Allowed operations
- Data access limits
Monitoring
Track MCP usage:
- Tool invocations
- Success/failure rates
- Response times
Security Considerations
Authentication
- MCPs use secure credentials
- Never expose secrets to AI
- Use scoped API keys
Data Access
- Limit to necessary data
- Use read-only when possible
- Audit access logs
Sandboxing
- MCPs run in isolation
- Cannot access other resources
- Rate limited by default
Creating MCPs
Build your own MCP servers:
- Define your tools
- Implement handlers
- Deploy the server
- Register with the platform
See Building MCP Servers for details.
Widgets
MCPs can power embeddable widgets:
- Chat interfaces
- Data displays
- Interactive tools
See MCP Widgets for details.
Next Steps
- Building MCP Servers - Create your own
- MCP Widgets - Embeddable components
- API Reference - Technical docs