# Connecting an AI Agent

> Source: https://docs.trailspark.ai/docs/connect-an-agent

## Overview

A **connected agent** links to your workspace through a sign-in-and-approve flow — the same way you'd connect any app to an account. There's no key to copy: you point the agent at your workspace's endpoint, and TrailSpark asks you to sign in and approve exactly what the agent can do.

This is the best fit for tools like **claude.ai**, **Claude Code**, and **Cursor**, which walk a person through connecting.

## Connecting an agent

1. Go to **Settings** > **Agent Access** and find the **Connect your agent** card.
2. Copy the **MCP endpoint** shown there (use the copy button). It looks like `https://app.trailspark.ai/mcp`.
3. In your agent, add TrailSpark as a connection and paste in the MCP endpoint. The exact spot varies by tool — the **Connect your agent** card links to short guides for connecting from claude.ai, Claude Code, and Cursor.
4. The agent sends you to the TrailSpark **approval screen** (see below). Approve it, and the agent is connected.

> [!NOTE]
> **MCP** (Model Context Protocol) is the open standard these agents use to talk to tools like TrailSpark. You don't need to know anything about it beyond copying the endpoint — the agent handles the rest.

## The approval screen

When an agent asks to connect, TrailSpark shows an approval screen titled **"[Agent] wants to access your Trailspark workspace."** Here you decide exactly what it gets:

- **Workspace** — the organization the agent will access. If you administer just one, it's filled in for you; if you administer several, choose one.
- **Permissions** — check any combination of **Read**, **Configure**, and **Data corrections**. See [Agent Access Overview](/docs/agent-access-overview#permissions) for what each allows.
- **Allow this agent to trigger account re-evaluation (billable)** — off by default. When on, set a **Daily re-evaluation limit** to cap how many accounts it can re-score per day.

Click **Approve** to connect the agent, or **Cancel** to reject the request.

> [!NOTE]
> Only organization admins can approve agent access. If you don't administer any workspace, the screen asks you to have an admin connect the agent instead. Approving sends the agent back to where it came from with access to the workspace you chose.

## Managing connected agents

Connected agents appear under **Connected agents** on the **Agent Access** page. Each card shows the agent's name, its permissions, when it connected, and when it was last used.

To disconnect one, click **Revoke** on its card and confirm.

> [!WARNING]
> Revoking is immediate. Any connected sessions for that agent stop working right away. The agent would have to be approved again to regain access.

## Agent keys vs. connected agents

Connected agents and [agent keys](/docs/agent-keys) do the same thing — give an agent scoped access to your workspace — but connect differently. Use a connected agent when a person can approve access in a browser; use an agent key when the agent runs unattended and you'd rather paste a credential. See [Agent Access Overview](/docs/agent-access-overview#choosing-between-an-agent-key-and-a-connected-agent) for a side-by-side comparison.