Learn how to build MCP integrations that enable users to access tools on external third-party MCP servers and use those tools as agentic actions.
What are MCP Integrations?
MCP (Model Context Protocol) integrations connect Cortex Agentic AI to external MCP servers, automatically discovering and exposing their tools as agentic system actions. Unlike traditional integrations that implement specific business logic, MCP integrations are generic bridges that:
Connect to a vendor’s MCP server endpoint.
Authenticate using the vendor’s preferred method.
Dynamically discover available tools at runtime.
Execute tools on demand with user-provided arguments.
When an MCP integration is configured in Cortex:
The
list-toolscommand is executed periodically to discover available tools from the MCP server.Agentic actions are auto-generated for each discovered tool.
When an agent calls one of these actions, the integration’s
call-toolcommand is invoked with the tool name and parameters.
Integration development and existing integrations
If this is your first time contributing to the Cortex Platform content repository, review the general contribution guide first. This guide assumes you’re familiar with standard content contribution practices and the development workflow.
We recommend reviewing the following existing MCP integrations as examples:
GitHub MCP: Packs/GitHubMCP/Integrations/GitHubMCP
Generic MCP: Packs/GenericMCP/Integrations/GenericMCP
Atlassian Cloud MCP: Packs/AtlassianCloudMCP/Integrations/AtlassianCloudMCP
Cloudflare MCP: Packs/CloudFlareMCP/Integrations/CloudFlareMCP