MCP Servers

The Model Context Protocol (MCP) extends AI capabilities with external tools. Connect MCP servers to give your AI access to specialized functions like web search, file operations, API integrations, and more.

What is MCP?

MCP is a standard protocol that lets AI models interact with external tools and services. MCP servers provide:

• Tools - Functions the AI can call (search, calculate, fetch data)
• Resources - Data the AI can access (files, databases, APIs)
• Prompts - Pre-built prompt templates

Server Types

Catalyst supports three types of MCP servers:

STDIO

Local processes that communicate via standard input/output. Good for:

• Local development tools
• File system access
• Command-line utilities

HTTP

Web-based servers accessed via HTTP requests. Good for:

• Remote services
• Cloud-hosted tools
• REST API integrations

SSE (Server-Sent Events)

Real-time streaming connections. Good for:

• Live data feeds
• Long-running operations
• Real-time updates

Adding an MCP Server

1. Navigate to MCP Servers from the sidebar menu
2. Click Add Server
3. Configure the server:
   • Name - A friendly name
   • Type - STDIO, HTTP, or SSE
   • Configuration - Connection details (varies by type)

HTTP Server Example

Name: My Web Search
Type: HTTP
URL: https://mcp.example.com

STDIO Server Example

Name: Local Tools
Type: STDIO
Command: npx @modelcontextprotocol/server-filesystem

Authentication

MCP servers may require authentication:

Bearer Token

Enter your API token in the authentication section. The token is sent with each request.

HTTP Basic Auth

Provide a username and password for basic authentication.

OAuth

For servers requiring OAuth:

1. Click Setup OAuth on the server page
2. You’ll be redirected to authorize access
3. Tokens are stored and refreshed automatically

Using MCP Tools

In Chat

1. Open a chat session
2. Click the Tools dropdown in the input area
3. Select tools from your connected MCP servers
4. The AI can now use those tools when responding

In Prompts

1. Open a prompt template
2. Click Add MCP Tool in the tools panel
3. Select tools to make available
4. Tool information is included in the template context

Server Status

The server list shows connection status:

• Active - Connected and ready
• Inactive - Not connected
• Error - Connection failed
• Connecting - Establishing connection

Click Initialize to connect to an inactive server.

Testing Tools

Test tools before using them in production:

1. Open the server detail page
2. Click Test on any tool
3. Enter test parameters
4. View the tool’s response

Managing Servers

Disconnect

Click Disconnect to close the connection to a server. You can reconnect later.

Health Check

Use the health check to verify a server is responding correctly.

Delete

Remove a server and all its stored configuration.

Popular MCP Servers

Some commonly used MCP servers:

• Filesystem - Read and write local files
• GitHub - Interact with GitHub repositories
• Slack - Send messages and read channels
• Google Drive - Access Google Drive files
• PostgreSQL - Direct database queries

Check MCP Server Registry for more options.

Troubleshooting

Server Won’t Connect

• Verify the URL or command is correct
• Check authentication credentials
• Ensure the server is running
• Check network/firewall settings

Tools Not Appearing

• Refresh the server connection
• Check server logs for errors
• Verify the server supports tool discovery

OAuth Errors

• Re-authorize the connection
• Check token expiration
• Verify OAuth client configuration