MCP Servers

MCP Servers allow AI systems in Karyam to securely interact with external tools, services, and data sources.

Built on the Model Context Protocol (MCP), they provide a standardized way to connect AI models with real-world capabilities without requiring custom integrations for every application.

MCP Servers extend what an Agent or AI Flow can do beyond reasoning by giving them access to external systems.

Why MCP Servers Exist

Large Language Models can reason and generate responses, but they cannot:

Read company databases
Access GitHub repositories
Query monitoring systems
Execute infrastructure commands
Call internal APIs
Retrieve live business data

MCP Servers bridge this gap by exposing external capabilities as reusable tools.

How It Works

An Agent or AI Flow invokes a tool provided by an MCP Server whenever external information or actions are required.

User Request
      ↓
Agent / AI Flow
      ↓
MCP Server
      ↓
External System
      ↓
Result
      ↓
Model Response

This allows AI systems to combine reasoning with real-world actions.

What Can MCP Servers Connect To?

MCP Servers can expose tools from virtually any external system.

Examples include:

GitHub
PostgreSQL
Redis
Elasticsearch
AWS Services
Internal APIs
Monitoring platforms
Developer tools
Business applications

Organizations can connect existing infrastructure without changing how Agents or AI Flows are built.

Tool Discovery

Each MCP Server publishes one or more tools.

When connected to Karyam, these tools become available for use by Agents and AI Flows.

Example:

GitHub MCP Server

├── Create Issue
├── Search Repository
├── List Pull Requests
└── Get File Contents

The AI system chooses the appropriate tool during execution based on the user’s request and available permissions.

Authentication

MCP Servers support multiple authentication mechanisms depending on the external service.

Common authentication methods include:

API Keys
Bearer Tokens
OAuth
Custom Headers

Authentication is configured once when the server is connected, allowing tools to be used securely during execution.

Server Discovery

Karyam supports automatic tool discovery.

When an MCP Server is connected:

Connect Server
      ↓
Discover Available Tools
      ↓
Synchronize Tool Metadata
      ↓
Tools Available to Platform

This removes the need for manual tool registration and keeps tool definitions synchronized with the server.

Reusable Infrastructure

An MCP Server is shared infrastructure.

A single connected server can be reused by multiple AI systems.

MCP Server
     ├── Agent A
     ├── Agent B
     ├── AI Flow A
     └── AI Flow B

This promotes consistency, centralized management, and reduced duplication.

MCP Servers and Execution

During execution, an Agent may decide that external information is required.

Example:

User:
"Create a GitHub issue for this bug."

        ↓

Agent

        ↓

GitHub MCP Tool

        ↓

Issue Created

        ↓

Confirmation Returned

The Agent focuses on reasoning while the MCP Server performs the external action.

Security and Governance

MCP Servers operate within Karyam’s governance model.

Tool execution can be:

Logged
Audited
Restricted by permissions
Subject to approval workflows

This ensures external integrations remain secure and compliant with organizational policies.

Relationship to Other Concepts

MCP Server
      ↓
Tools
      ↓
Agent / AI Flow
      ↓
Run
      ↓
Logs

MCP Servers provide the integration layer that enables AI systems to interact with external services.

The Karyam Philosophy

AI should not operate in isolation.

It should seamlessly interact with the tools and systems your organization already relies on.

MCP Servers make those integrations standardized, reusable, and secure.