Integrations

What Cortex connects to, and how far along each one is.

Cortex reaches your systems through MCP, the open protocol for giving agents tools, and reaches models through one gateway. Its agents ask for a kind of model, never a particular vendor’s, so the model behind them is your choice. Each entry below says plainly where it stands; see the platform for what an agent does with them, and the recorded demos to watch it.

What the statuses mean

Available
Cortex has a connector for it today, and it works against the real service.
In the demos
You can watch it in the recorded demos, which run on simulated data. It is not a connector to the real service yet.
Any MCP server
Cortex connects to it the same way it connects to any MCP server.
Planned
On the Cortex roadmap. Not built yet.

Statuses reflect Cortex as of November 11, 2026, and change here when the work does.

CRM

  • SalesforceIn the demosThe recorded demos read accounts, contacts, opportunities and activity, and log a call or update an opportunity after you approve it, all on simulated data.

Support

  • ZendeskIn the demosThe recorded demos search tickets, read their history and satisfaction ratings, and sum up an organization’s support, on simulated data.

Engineering

  • JiraIn the demosThe recorded demos look up the issues behind a customer’s tickets and whether their fixes shipped, on simulated data.

Calls

  • ZoomIn the demosThe recorded demos list a customer’s meetings and read their transcripts for who said what and what was promised, on simulated data.

Productivity suites

  • Google WorkspacePlannedSign-in with your Google Workspace accounts through standard OIDC. This covers who can use Cortex, not reading Gmail, Drive or Calendar.
  • Microsoft 365PlannedSign-in with Microsoft Entra ID, the identity behind Microsoft 365, through standard OIDC. This covers who can use Cortex, not reading Outlook, Teams or SharePoint.

Model providers

  • ClaudeAvailableAnthropic’s Haiku, Sonnet and Opus models, called through the gateway with an Anthropic API key.
  • GeminiAvailableGoogle’s Gemini Flash, Flash-Lite and Pro models, called through the gateway with a Gemini API key.
  • Vertex AIAvailableGemini on Google Cloud, called with the deployment’s own service account rather than a key. A Cortex deployment can run on Vertex AI alone.
  • Local modelsAvailableModels on your own hardware, served by Ollama or vLLM, for chat, embeddings and search ranking.
  • Amazon BedrockPlannedA deployment that runs on Bedrock alone, tested end to end, is on the roadmap. It has not been run yet.

MCP

  • Any MCP serverAny MCP serverCortex connects to servers that speak the Model Context Protocol, over HTTP or as a local process. An administrator can add one for everyone, each person can add their own, and an agent uses a server’s tools only when its definition allows it.