AI & Agentic Overview

DataZen extends beyond classic data integration into AI and agentic workflows. The same governed pipelines that move data can call external agents and process responses, be exposed as MCP Tools, and be administered by agents through the DataZen API, while assistants can author SQL CDC and cloud functions that pipelines invoke.

Four complementary usage patterns cover the main scenarios:

Diagram of four DataZen AI agent use cases: Vibe Coding, Pipeline to Agent, Agent to Pipeline, and Agent to DataZen API — Four ways AI agents work with DataZen: generating scripts and cloud functions, calling agents from pipelines, calling pipelines as MCP tools, and administering the agent via the DataZen API.

Use case	What it does	npm package / capability
Vibe Coding	Build SQL CDC scripts with your favorite agent, and/or generate external HTTP cloud functions that pipelines can call later. For script generation, agents use the SQL CDC language specification via MCP.	`datazen-sqlcdc` (SQL CDC generation; cloud functions are built outside DataZen)
Pipeline → Agent	A running pipeline invokes an LLM through an AI Endpoint using `CALL AGENT` or `APPLY HTTP`—for classification, enrichment, summarization, or other inline decisions—without requiring a separate MCP package. And the pipeline can also process the response if desired to continue its processing.	`CALL AGENT` or `APPLY HTTP` (no npm package)
Agent → Pipeline	Expose Reader or Direct pipelines as MCP Tools (Pipelines as Tools). Develop and test them in Portal AI Chat, then connect Claude, Hermes, Cursor, and other agents to live pipeline results.	`datazen-pipeline-mcp`
Agent → DataZen API	Let agents call the DataZen HTTP API as MCP tools for cloud and self-hosted agents—health, jobs, history, variables, and start/stop/activate operations (DataZen API Tools).	`datazen-api-mcp`

The three MCP servers are published under the Enzo Unified npm organization. Browse the packages here: npmjs.com/settings/enzounified/packages. For client credentials and environment variables, see MCP Settings.

Vibe Coding

Use your preferred coding agent to generate and refine integration artifacts. With datazen-sqlcdc, the assistant has first-class access to the SQL CDC language specification (list_operations, get_operation, get_examples) so it can produce correct scripts instead of guessing. You can also have AI generate external HTTP cloud functions (or similar endpoints); once deployed, a pipeline can invoke them over HTTP just like any other remote service.

No agent credentials are required for the language-spec MCP server. See AI-Generated SQL CDC for package setup and examples.

Pipeline → Agent

Pipelines can call an LLM mid-flight without exposing the pipeline as a tool. Configure an HTTP/S connection as an AI Endpoint, then use the SQL CDC CALL AGENT operation to send a prompt (and optional context from the current row or pipeline state) and receive a model response inline. That response can drive filters, branches, new columns, or downstream writes.

This pattern does not require an npm package: runtime is entirely inside DataZen. See CALL AGENT and HTTP Connection: AI Endpoint.

Agent → Pipeline

Turn governed DataZen pipelines into tools an LLM can call (Pipelines as Tools). Configure an AI Endpoint, define MCP Tool metadata on a Reader or Direct pipeline, validate behavior in Portal AI Chat, then register datazen-pipeline-mcp in an external agent. Optional DECLARE parameters become tool inputs. See Pipelines as Tools and MCP Tool.

Agent → DataZen API

Operate the agent control plane from an LLM (DataZen API Tools): ping and status, list jobs, read history and logs, inspect variables and connection metadata, and start, stop, or activate pipelines. Works for both cloud and self-hosted deployments via datazen-api-mcp and a Service Token. See DataZen API Tools.

How They Fit Together

Use Vibe Coding to author SQL CDC scripts and optional cloud functions that pipelines will run.
Use Pipeline → Agent when a pipeline needs an LLM decision or enrichment during execution (CALL AGENT).
Use Agent → Pipeline to let external agents query or act through governed pipelines as MCP tools.
Use Agent → DataZen API when the agent must manage or observe the DataZen agent itself.