AI Agents¶
Use this section when you are connecting Centrali to Claude, Cursor, Codex, Windsurf, or any other MCP-capable assistant.
Choose a Path¶
| If you need to... | Start here | Why |
|---|---|---|
| Get connected fast in an interactive AI client | AI Agent Quick Start | Fastest setup path with the least context switching |
| Configure the hosted or stdio MCP transport directly | MCP Server guide | Canonical connection and auth details |
| Debug auth, routing, tool failures, or stale examples | AI Agent Troubleshooting | Best place for common failure modes and host rules |
| Understand what the SDK itself can do outside MCP | SDK Overview | Useful when an agent workflow turns into direct app code |
Start Here¶
- New setup: Follow the AI Agent Quick Start for the fastest path.
- Connection details: Use the MCP Server guide for hosted and stdio configuration.
- Debugging: Use AI Agent Troubleshooting when tools, auth, or trigger execution do not behave as expected.
Which Path Should You Use?¶
Hosted MCP Server¶
Use this for interactive AI clients such as Claude Desktop, Claude Code, and Cursor. It is the default path for most users because authentication happens in the browser and you do not need to manage service-account secrets locally.
Stdio MCP Package¶
Use this for CI jobs, headless automation, or environments where browser login is not available. This path uses a service account and explicit environment variables.
What AI Agents Can Do¶
- Inspect collections and records
- Create or update records
- Manage compute functions and triggers
- Work with pages, automations, and IAM resources
- Use schema-discovery and troubleshooting tools to orient themselves before taking action
Start with describe_centrali, then move into the specific describe_* tools for the domain you need.