# Ingestion your AI agents can run

> Ask an agent to investigate a failure, create a connection or run a job. Through the Erathos MCP server, through a coding agent that builds connectors for you, or through the API, straight from your code.

Source: https://www.erathos.com/en/ai
Em português: https://www.erathos.com/ai

## Talk to your workspace

The Erathos MCP server lets an AI agent, like Claude, act on your workspace from a conversation: list connections and jobs, investigate a failed run, choose what to sync and trigger runs, without going through the UI for every step.

The agent only acts within the workspaces your account can already access, and every call is checked against your real permissions. Credentials stay in workspace secrets, never in the conversation.

### Diagnose failures

Ask why a job failed and the agent brings back the run's logs, timeline and metrics.

### Set up connections and jobs

Create connections and destinations and choose tables and sync options. The agent shows a preview before saving.

### Orchestrate the next step

Trigger runs and backfills, and set up webhooks that start Airflow or dbt Cloud when a job finishes.

## Your connector, built by a coding agent

The custom connector template ships with a skill for coding agents. Open the repository in Claude Code, describe the API you want to connect, and the agent builds the connector for you, with no YAML written by hand.

It calls the real API to work out fields, primary key and pagination, flags what it could not verify, and runs the validation until the connector passes. Every push is a deploy, so it never pushes without asking.

### Describe the API

Say what you need, like "sync orders and customers incrementally", and the agent asks the rest: authentication, pagination and filters.

### Built from real responses

Fields, response path and primary key come from real API calls, not from guesses based on the docs.

### Validated before the push

The agent runs the validation until it passes and only pushes with your approval. The connector shows up in the catalog with a Custom badge.

## Everything you do in the platform, from code too

The developer API covers the whole setup: create connections, configure jobs, create destinations and run jobs. Provision ingestion from your own product, from scripts or from your CI pipeline.

Jobs get a preview before you save, and each connection reports which setup step it is on, so your code knows what to do next.

### Create connections and destinations

Pick the connector, send the credentials or authorize through OAuth, and follow setup_status until the connection is ready.

### Configure jobs with a preview

See tables, endpoints and sync options, check the preview, then save with dry_run: false.

### Run and follow

Trigger jobs and follow every run through the API, with the same key as the endpoints you already use.

## Frequently Asked Questions

### What is MCP, and which tools does it work with?

MCP (Model Context Protocol) is an open standard for connecting AI agents to other systems. The Erathos MCP server works with MCP-compatible clients such as Claude: you add the server to your client and authenticate with your Erathos account. The steps are in the [MCP docs](https://docs.erathos.com/platform/mcp).

### Does the agent see my credentials?

No. To create a connection, the agent hands you a link to enter the credentials in the platform itself, and they stay in workspace secrets, never in the conversation. If whoever holds the credentials has no Erathos account, the agent can create a single-use secure link for them.

### Can the agent do anything I can't?

No. It only acts within the workspaces your account can already access, and every call is checked against your permissions at the time of the request.

### Which plans include custom connectors?

Custom connectors are on the Pro and Enterprise plans. The coding-agent skill ships with the connector template: Claude Code loads it automatically, and other agents can be pointed at the skill file. See the [custom connector docs](https://docs.erathos.com/connectors/custom-connectors).

### How do I authenticate with the API?

With an Erathos API key, sent in the Authorization header. The key acts as the user who owns it: each request can only do what that user can already do in the workspace.