> For the complete documentation index, see [llms.txt](https://docs.erathos.com/llms.txt). Markdown versions of documentation pages are available by appending `.md` to page URLs; this page is available as [Markdown](https://docs.erathos.com/platform/mcp.md).

# MCP Server

Let AI agents interact with your Erathos workspace.

## Overview

Erathos exposes a [Model Context Protocol](https://modelcontextprotocol.io) (MCP) server, letting an AI agent (like Claude) act on your Erathos workspace directly from a conversation — checking on connections and jobs, diagnosing a failed run, finding which field or table holds a piece of data, or setting up a new connection, without switching over to the platform's own UI for every step.

* [**How to Connect**](/platform/mcp/how-to-connect.md) — add the Erathos MCP server to your AI client and authenticate.
* [**Available Tools**](/platform/mcp/available-tools.md) — the full reference for everything the server currently exposes.

## What you can do with it

* **Discover your workspace** — list your workspaces, connections, jobs, and runs; find which endpoint or table field holds a piece of data.
* **Diagnose issues** — pull the full detail on a failed run: its logs, execution timeline, and per-model metrics.
* **Set up connections** — create a new source connection or destination; the agent hands you a link back to the platform to finish entering credentials, since it never asks for them directly. If someone without an Erathos account holds the credentials, the agent can create a single-use secure link for them instead. It can then choose the connection's settings (e.g. which ad account it reads from).
* **Configure jobs** — choose which tables or endpoints sync and how, across a whole database in a few steps. You say how fresh the data must be and whether you need each row's latest version or its history of changes; the agent picks the rest, and shows you a preview before saving anything.
* **Manage jobs** — trigger or cancel runs (including full refreshes and custom backfills), and enable or disable jobs.
* **Orchestrate what comes next** — set up webhooks that start your next step elsewhere (an Airflow DAG, a dbt Cloud job) after a job or a whole connection syncs, with rules for when they fire; and check what each webhook did after each run, to see why one didn't fire or whether it's failing. Credentials stay in workspace secrets, never in the conversation.
* **Get human help** — when the agent can't solve a problem, it can open a ticket with Erathos support for you, with the run, job and connection details already filled in. You follow it in the platform's chat.

{% hint style="info" %}
The agent only ever acts within the workspaces your own Erathos account already has access to — every tool call is checked against your real permissions, live, on every request.
{% endhint %}