Celigo's July release: a headless platform that agents can run

For most of their history, iPaaS products have been built around a visual designer: you open the browser, drag steps onto a canvas and click through settings. Celigo’s July 2026 release shows another direction. Much of the platform can now be run without the interface at all, from a terminal, from a developer’s IDE or by an AI agent.

What Celigo released

The release, published on 6 July, is generally available. The main parts are:

• A headless platform. A new developer portal gathers the REST API documentation, a command-line tool and Platform MCP, a hosted MCP server that lets AI agents manage integrations through conversation. The Celigo CLI gives full access to the platform from a terminal or IDE.
• Claude as a third model. Celigo’s AI agents can now use Anthropic’s Claude in addition to OpenAI and Gemini, with tool calls, structured output, reasoning in several steps and guardrails. For now, customers must bring their own API key.
• Data ingestion. A new built-in component loads data from SaaS applications, databases and ERP systems into data warehouses, using the same connectors and governance as the rest of the platform.
• Faster databases. Bulk loading into SQL Server and BigQuery is up to five times faster, according to Celigo, and PostgreSQL gets real-time change data capture.
• Celigo Storage gives platform-managed file storage, so an external SFTP server or S3 bucket is no longer required.

Why it matters

Platform MCP is the most interesting feature. Most MCP work in iPaaS so far has been about letting agents call integrations: an integration becomes a tool the agent uses. Platform MCP goes one level up. It lets an agent build, change and troubleshoot the integrations themselves.

That can save a lot of time, but it also changes the risk. An agent that can change integration flows can break them as well. The same rules that apply to people should apply to agents: separate environments, approval before production and a log of who changed what.

The data ingestion component is also worth noting. Celigo has mostly been known for process integration between applications. Moving data into a warehouse has usually been a job for separate ELT tools. Having both in one platform means fewer tools to buy and govern.

How it compares

Several iPaaS vendors are moving in the same direction: CLI tools, MCP servers for the platform itself and developer tools inside the AI assistants people already use. Celigo’s version stands out for being hosted and generally available, and for covering both integration and data ingestion.

What to ask

1. What may an agent change? Decide whether Platform MCP should be allowed in production, or only in development and test environments.
2. Is every change logged? Check that changes made through the CLI and MCP appear in the same audit trail as changes made in the interface.
3. Do you still need a separate ELT tool? Compare the new ingestion component with the tool you use today, for the sources you actually have.