Connect ChatGPT

Install the Orcha plugin in ChatGPT, or add Orcha as an MCP connector. Then chat, deep research, and company knowledge can search and cite your workspace.

Connection methods

With both methods, ChatGPT connects to Orcha over MCP and OAuth. The Orcha server is at https://app.tryorcha.com/mcp and uses Streamable HTTP.

The server registers the ChatGPT client through dynamic client registration with PKCE. Thus, you do not copy a client ID or a secret. After the connection, ChatGPT signs you in to Orcha. You then select the workspace and the permissions for the connection.

The Orcha plugin. ChatGPT and Codex use the same OpenAI plugin directory. The Orcha plugin contains the MCP server and the Orcha usage guides as skills. The skills tell the model which tool answers each type of question. After installation, the plugin runs a setup skill. Use the plugin when the directory lists it. Until then, use the connector. Also use the connector if your workspace blocks the directory.
A custom connector. You add the same MCP server by its URL. This method works now on all plans that allow custom connectors. It does not need a directory listing.

With both methods, ChatGPT calls Orcha in one of two modes:

Standard mode. Chat, deep research, and company knowledge use only two tools, search and fetch. Orcha implements both tools to the ChatGPT contract. search returns the documents that match, each with a link. fetch returns one full document. ChatGPT cites results by those links.
Developer mode. When you turn on developer mode in the ChatGPT settings, the connection gives access to all 30 Orcha tools. These include search_context, memories, structured databases, and the write tools. ChatGPT asks for confirmation before it runs a tool that writes.

Prerequisites

You must have a ChatGPT plan that allows plugins or custom MCP connectors. OpenAI controls this by plan and by workspace policy. On Business, Enterprise, and Edu, an admin possibly must enable them first. If the option is not available, examine the current OpenAI plan table.
You must be a member of the Orcha workspace that you want to connect.

Install the plugin

In ChatGPT, open the Plugins Directory.
Search for Orcha.
Select the plus button to install the plugin. ChatGPT opens the Orcha sign-in.
Sign in.
On the consent screen, select the workspace.
Select the permissions for the plugin: read, write, and delete. The default is read only.
Start a new chat.

The setup skill of the plugin confirms the connected workspace. It gives a summary of the workspace content. Then it runs a first search with citations.

To install the plugin in Codex:

Run /plugins.
Search for Orcha.
Install the plugin with the same procedure.

You can also install the plugin directly from its public repository. You can do this before the directory lists the plugin, or instead of the directory:

codex plugin marketplace add westonhancock/orcha-mcp

Then:

Open /plugins.
Select the orcha marketplace.
Install the plugin.

Add the connector instead

In ChatGPT, open Settings.
Open Connectors. Some builds show this item as Apps.
Select Create.
Type the name Orcha.
Type the MCP server URL:
https://app.tryorcha.com/mcp
Set the authentication to OAuth.
Keep the client ID and secret empty. Orcha registers the ChatGPT client at the first contact.
Create the connector. ChatGPT opens the Orcha sign-in.
Sign in.
On the consent screen, select the workspace for the connector.
Select the permissions for the connector: read, write, and delete. The default is read only.

Your workspace role at consent time sets the maximum permissions. Orcha checks the permissions again on each request. If you lose a role, the connector loses those permissions immediately.

You can revoke the grant at any time on the Orcha MCP settings page. To change the workspace, reconnect.

Use it

Enable the Orcha connector in a conversation or in a deep research run. Then ask for the information that you need:

"Search my Orcha workspace for our deployment requirements"
"Using Orcha, summarize what the team decided about pricing"
"Run deep research on our onboarding flow using the Orcha connector"

Each result has a link to the file in Orcha. For content from a connected source, the link goes to the upstream document. ChatGPT cites the results.

In developer mode, you can also tell ChatGPT to load a memory, query a structured database, or create a file. For the full tool list, refer to the MCP reference.

What search and fetch return

search runs the Orcha hybrid retrieval on retained content only. It combines the passages that match into one result for each document. The results are Orcha files, structured database records, and documents that Orcha syncs from connected sources. Each result has an id for fetch, a title, a url, and a short excerpt.

search does not contact upstream providers. Deep research sends one question as many searches, and each live call uses the daily budget of the connection. For live results, use developer mode and search_context with mode: "fresh".

fetch takes the id from a search result. It returns the full text of the document with its link and metadata. Retained connected-source documents come from the index. If a document is in the catalog but Orcha does not retain it, fetch gets it live from the provider. This live fetch uses the daily budget.

A folder-scoped grant sees only its folders through both tools. It never sees connected-source documents.

Troubleshooting

The plugin is not in the directory. Use the connector procedure above. Or, in Codex, install the plugin from the repository marketplace. Both methods connect to the same server with the same permissions.
ChatGPT reports that the server has no search or fetch tool. You created the connector against an earlier Orcha deployment. Delete the connector and create it again.
invalid_client after a long time without use. ChatGPT removed its registered client. Delete the connector in ChatGPT and create it again.
Results are not current. search reads retained content. Start a sync on the source in Orcha. Or turn on developer mode and use search_context with mode: "fresh".
Write tools are not available. You must use developer mode in ChatGPT. The grant must also include write. To change the grant, reconnect.