Skip to main content

Connectors

Connectors are services that implement the Model Context Protocol, providing tools and resources that AI can use. Understanding how connectors work is essential for extending Obot's capabilities and integrating with external systems.

MCP Servers​

What is an MCP Server?​

An MCP Server is a service that implements the Model Context Protocol specification. It can provide:

  • Tools: Functions that agents can call to perform actions
  • Resources: Data that agents can read (files, databases, APIs)
  • Prompts: Pre-defined prompt templates for specific tasks
  • Sampling: Custom completion endpoints for specialized models

Server Types​

Hosted Services​

External services that implement MCP natively:

  • Cloud APIs: Services like GitHub, Slack, Google Drive
  • SaaS Platforms: CRM systems, project management tools
  • AI Services: Image generation, translation, analysis services
  • Data Providers: Weather APIs, news feeds, financial data

Self-Hosted Servers​

MCP servers you run in your own infrastructure:

  • Database Connectors: PostgreSQL, MySQL, MongoDB interfaces
  • File System Access: Local or network file system tools
  • Internal APIs: Your organization's custom services
  • Legacy System Bridges: Connect to existing enterprise systems

Containerized Servers​

MCP servers deployed as containers in Obot's infrastructure:

  • Secure Isolation: Runs in controlled environments
  • Network Security: Restricted network access policies

Server Configuration​

Connection Settings​

  • Endpoint URLs: Where to connect to the MCP server
  • Authentication: How to authenticate with the server

Capability Declaration​

MCP servers declare their capabilities:

  • Available Tools: List of tools and their parameters
  • Resource Types: What kinds of data can be accessed
  • Required Permissions: What access the server needs
  • Version Information: Protocol and server version details

Tools​

Tool Definition​

Tools are specific functions that MCP servers expose for AI agents to use. Each tool has:

  • Name: Unique identifier for the tool
  • Description: What the tool does and when to use it
  • Parameters: Required and optional inputs
  • Output Schema: Format of the tool's response
  • Error Handling: How errors are reported and handled

Tool Execution​

Request Flow​

  1. Tool Discovery: Agent discovers available tools from MCP server
  2. Parameter Validation: Agent validates required parameters
  3. Authentication: Verify permissions and credentials
  4. Execution: MCP server executes the tool function
  5. Response Processing: Handle results and potential errors

Error Handling​

  • Parameter Errors: Invalid or missing parameters
  • Authentication Errors: Insufficient permissions or invalid credentials
  • Service Errors: External service unavailable or failing
  • Timeout Errors: Operations taking too long to complete
  • Rate Limit Errors: Too many requests in a given time period