> ## Documentation Index
> Fetch the complete documentation index at: https://archie.com/docs/llms.txt
> Use this file to discover all available pages before exploring further.

# Archie MCP

> Agent Access lets AI assistants like Claude, ChatGPT, and Cursor read your Archie projects. You turn it on for your account, and you approve each client in your browser.

**Agent Access** is the account-level MCP connection. When it is on, an MCP client can ask to read your Archie projects, and you approve that client in your browser. Approved clients act as you and reach every project you can.

It is separate from the per-project [MCP Server](https://archie.com/docs/mcp/mcp-server.md), which you enable for one project's data.

## Find Agent Access

The **MCP** button sits in the header, to the left of the notifications bell. It appears on the organization dashboard and inside every project.

![The MCP button in the header of a project](https://mintcdn.com/archie-e998dbf6/Lv07l3TYoPUeWEyf/images/46-mcp-header-button-project.png?fit=max\&auto=format\&n=Lv07l3TYoPUeWEyf\&q=85\&s=690bcf48cf2a77e77df5efb5f9692b6f)

Click **MCP** to open the Agent access panel. Press **esc** or click away to close it.

![The Agent access off panel, showing the assistants Archie works with and the Turn on Agent Access link](https://mintcdn.com/archie-e998dbf6/Lv07l3TYoPUeWEyf/images/44-mcp-header-button.png?fit=max\&auto=format\&n=Lv07l3TYoPUeWEyf\&q=85\&s=758374a37c553b70794405cd26b4cafb)

When Agent Access is off, the panel is titled **Agent access off**. It lists the assistants it **Works with**: ChatGPT, Claude, Gemini, Copilot, Cursor, and Perplexity. The **Turn on Agent Access** link opens **Settings → Security**.

## Turn it on

**1. Open Security settings**

Click **MCP** in the header and choose **Turn on Agent Access**, or open your account settings and go to **Security**.

**2. Switch on Agent Access**

Use the **Agent Access** switch. Archie then publishes an **Endpoint** on the same card.

![The Agent Access card in Security settings with the switch off](https://mintcdn.com/archie-e998dbf6/Lv07l3TYoPUeWEyf/images/45-mcp-agent-access-card.png?fit=max\&auto=format\&n=Lv07l3TYoPUeWEyf\&q=85\&s=5c0a22e3a8ee628cb01dd7860293a9d2)

## Connect a client

**1. Copy the endpoint**

Click **Copy** next to **Endpoint**, either in the Agent access panel or on the **Security** page. The endpoint is not a secret, so you can share it with your team.

**2. Add it to your assistant**

Point your assistant at the endpoint. See [Connecting an assistant](https://archie.com/docs/mcp/connecting.md) for where each tool takes a URL.

**3. Approve the client**

On its first call, the client sends you to your browser. Approve it there.

## Allow agents to write

By default, agents can only read. When Agent Access is on, the card shows an **Agents may write** switch. Turning it on adds three write tools for every project at once. Turning it off removes the write tools from the manifest.

> [!WARNING]
>
> Approved clients act as you. Leave **Agents may write** off unless you want an assistant to change your projects.

## Turn it off

Switch **Agent Access** off in **Settings → Security**, then confirm with **Turn it off**. Every client you approved loses access on its next call, across every project at once. To use Agent Access again, you approve each client again.

## Archie MCP vs. MCP Server

| | Archie MCP | [MCP Server](https://archie.com/docs/mcp/mcp-server.md) |
| - | - | - |
| **Scope** | Every project your account can reach | One project |
| **Turn it on in** | The header **MCP** button or **Settings → Security** | **Backend → Core Services → MCP Server** |
| **Sign-in** | You approve each client in your browser | The project's sign-in, or a project API key |
| **Acts as** | You | The connecting user, with their project roles |


This documentation is built and hosted on [Mintlify](https://mintlify.com), a developer documentation platform.