# Use with ChatGPT

Connect Archive to ChatGPT or Codex to search captured content, look up creators, and manage saved views from a conversation. Setup requires only the endpoint URL — there is no client ID, no token to copy, and nothing to install.

```http
POST https://app.archive.com/api/v2/mcp
```

> **ChatGPT custom connectors need developer mode**
>
> Adding your own MCP server in ChatGPT requires developer mode, which needs a paid ChatGPT plan and, on a workspace plan, an admin or owner to enable it and grant the role permissions. Write tools are gated further: OpenAI currently limits write and modify actions to Business, Enterprise and Edu workspaces, so a Plus or Pro connection may see the read tools only. Both the switch and those limits have moved more than once, so see [OpenAI’s guide](https://help.openai.com/en/articles/12584461-developer-mode-and-mcp-apps-in-chatgpt) for what applies to your account today.

## Set up your client

Follow the instructions for your client. After you authorize the connection, the client can read — and if you granted write access, update — content in the workspaces you can already access.

Add Archive as a developer-mode connector. ChatGPT’s deep research is read-only — it drives a `search`/`fetch` tool pair and never writes or modifies anything — and Archive’s tools are named for the product surface instead: `searchItems`, `searchCreators`, `getWorkspaces`.

1. Turn on developer mode for your account, as described above.
2. Add a custom connector, paste the endpoint URL above, and choose OAuth authentication.
3. Log in to Archive when the consent screen opens, and choose whether to grant the optional **write** scope.
4. Back in the chat, enable the connector in the tools menu.

> **Write access is granted at consent time**
>
> Without the write scope, the client sees only the read tools — `tools/list` omits the write half entirely rather than advertising tools that would fail. ChatGPT applies its own limits on top of that scope: whether it can call a write tool at all depends on your plan and workspace permissions, and it generally asks you to confirm each write before it runs.

## What happens during authorization

ChatGPT and Codex both connect with **OAuth 2.1**. You don’t create or paste an API token:

1. The client’s first call gets HTTP 401 with a `WWW-Authenticate` challenge naming Archive’s protected-resource metadata.
2. It fetches the authorization-server metadata and registers itself via Dynamic Client Registration — nothing to configure on Archive’s side.
3. You log in at Archive’s consent screen and choose the scopes to grant. Without the optional **write** scope the connection is read-only.
4. The client exchanges the authorization code for an opaque, Archive-issued access token scoped to what you approved.

Each person authorizes their own connection, and ChatGPT and Codex register separately, so revoking one leaves the others working. Scopes are chosen at consent time only, so to change them, remove the connector and add it again.

> **ChatGPT connects from OpenAI’s servers**
>
> The connection is made from OpenAI’s infrastructure rather than from your browser, so a `localhost` or private-network address is unreachable. Use the public endpoint above.

## Verify it works

Ask ChatGPT to list your Archive workspaces. If the connection works, it calls `getWorkspaces` and returns your actual workspace names and IDs. In Codex, `/mcp` lists the connected servers and the tools they expose. If the tool list is empty or every call returns 401, authorization didn’t complete — remove the connector and add it again.

See [Example prompts](/api/v2/docs/mcp/example-prompts) for more starting points, or [Troubleshooting](/api/v2/docs/mcp/troubleshooting) if something isn’t working.

Running a server-side agent instead? A bearer token from the Integration tab also works, but it is a write-capable, server-side secret rather than a per-person authorization. See [Connect a client](/api/v2/docs/mcp/connect) for when to use which.