Skip to content
Karyam

Server Discovery

One of the biggest advantages of the Model Context Protocol (MCP) is dynamic discovery.

Traditional integrations require developers to manually define:

  • API endpoints
  • Input parameters
  • Response formats
  • Authentication rules
  • Tool descriptions

With MCP, servers describe their own capabilities.

Karyam simply discovers and imports them.


Without discovery:

GitHub Integration
↓
Manually Define:
- Create Issue
- Create Pull Request
- List Repositories
- Search Code
- Get Workflow Runs

Every integration requires engineering effort.


With MCP:

Connect GitHub MCP Server
↓
Discover Tools
↓
Use Tools Immediately

No manual schema creation is required.


After connecting an MCP Server:

Save Server
↓
Authenticate
↓
Sync Tools
↓
Discover Available Capabilities
↓
Import Tool Definitions
↓
Tools Available To Agents & Flows

Navigate to:

MCP Server
↓
Tools
↓
Sync Tools

Karyam will connect to the MCP Server and request the list of available tools.


Suppose we connect a GitHub MCP Server.

After synchronization Karyam may discover:

GitHub MCP Server
│
├── Create Issue
├── Create Pull Request
├── Search Repository
├── Get Workflow Runs
├── List Repositories
├── Create Branch
└── Merge Pull Request

These tools immediately become available for use.


Every discovered tool includes metadata that can be viewed by expanding the tool.

Typical metadata includes:

  • Tool name
  • Description
  • Input schema
  • Required parameters
  • Optional parameters
  • Output schema

Example:

Create Issue
├── repository
├── title
├── body
└── labels

This allows users to understand exactly how the tool behaves before attaching it to Agents or AI Flows.


Example:

Tool:
Create Pull Request
Description:
Create a new pull request in a repository.
Inputs:
- repository
- branch
- title
- description
Output:
- pull_request_url
- pull_request_number

One of MCP’s strengths is that capabilities can evolve over time.

Example:

GitHub MCP Server v1
├── Create Issue
└── List Repositories
GitHub MCP Server v2
├── Create Issue
├── List Repositories
├── Search Code
└── Merge Pull Request

Running Sync Tools imports the new capabilities automatically.

No platform updates are required.


If an MCP Server changes:

  • Tool definitions
  • Input schemas
  • Available capabilities

simply run:

Sync Tools

Karyam refreshes the local metadata and updates the available tools.


Discovered tools can be attached directly to:

  • Agents
  • AI Flows

Example:

Agent Settings
↓
Attach MCP Tools
↓
Select:
- Create Issue
- Search Repository
- Merge Pull Request

Developer:
Create a bug report for failed VPN renewals.
Agent
↓
GitHub MCP Tool
↓
Create Issue
↓
Issue Created

Discovery does not automatically grant permission to execute tools.

Organizations can still control access using:

  • RBAC
  • Team permissions
  • Approval workflows
  • Human-in-the-Loop approvals

Example:

Merge Pull Request
↓
Approval Required
↓
Engineering Manager Approval
↓
Execute Tool

Karyam records:

  • Discovery attempts
  • Synchronization events
  • Tool metadata changes
  • Execution history
  • Errors

This provides full visibility into external capabilities available within the platform.


Verify:

  • The MCP Server is running.
  • Authentication is valid.
  • The connected account has permission to list tools.

Run:

Sync Tools

The server may expose new capabilities that have not yet been imported.


Verify:

  • Network connectivity
  • Server availability
  • Authentication configuration

Traditional integrations require developers to define capabilities manually.

MCP changes this.

Connect Server
↓
Discover Capabilities
↓
Build AI Systems

This dramatically reduces integration effort while increasing flexibility.


Continue with:

➡️ Examples

See real-world examples of how MCP Servers can be used inside production AI systems.