HowtoconnectChatGPTtolivefinancialdata

KEY TAKEAWAYS

There is more than one way in, and they are not interchangeable.

1. What you can connect is limited by your data licence, not by ChatGPT.

2. A connection that cannot be cited is a connection you cannot audit.

3. Set it up once for the team rather than once per analyst.

4. Custom GPTs are scheduled to retire on 11 December 2026, so build on plugins and MCP apps instead.

You can connect ChatGPT to live financial data using a provider’s plugin. Once the plugin is installed and you have signed in with your licensed account, ChatGPT can search the provider’s data when you ask a question. Firms can also connect their own data through custom MCP apps or use ChatGPT for Financial Services.

What are the ways ChatGPT can reach outside data?

ChatGPT has four working routes to outside data as of 30 September 2026, and a fifth that is on its way out. They are not interchangeable, and the vocabulary has shifted enough that older guides will mislead you.

The main difference is between an app and a plugin. OpenAI describes an app as a connection between ChatGPT and an external service, while a plugin brings apps and skills together in a reusable workflow. Some plugins contain only skills and do not connect to an outside service. What OpenAI previously called a connector is now called an app, although the older term may still appear in settings and vendor documentation.

The fifth option, custom GPTs, is being phased out. OpenAI plans to retire them on 11 December 2026, with a later deadline for Enterprise workspaces that receive an approved extension. Enterprise users are also expected to lose the ability to create new custom GPTs from 26 October. Custom actions that allow a GPT to call a data API will not move across automatically, so setups built with an API key and schema will need to be rebuilt as custom MCP apps. If you are starting a new financial data setup, a custom MCP app is therefore the more suitable route.

Which one suits a single analyst, and which suits a team?

For an individual analyst, the simplest option is to install the provider’s plugin and sign in with their own account. For a team, an administrator can install and manage the same plugin across the workspace. The data remains the same; the main difference is who manages access and permissions.

For an individual analyst, a plugin can only access the content available through their provider account. This keeps the same access limits already set by the provider. Before setting it up, it is worth checking whether the licence allows personal use. Pro users can also add MCP servers with read and fetch permissions through developer mode, but full MCP support, including write actions, is limited to Business, Enterprise and Edu.

For a team, an administrator can make a plugin available for members to install or add it to the workspace automatically. They can also review the access and actions included with each app. Enterprise and Edu administrators can turn individual app actions on or off. Business workspaces offer less flexibility because a published custom app cannot be edited and must be recreated when changes are needed. A shared setup also helps keep sources, instructions and access rules consistent across the team, rather than leaving each analyst to configure them separately.

What can it actually pull in once connected?

Once connected, ChatGPT can pull earnings transcripts, financial statements, company fundamentals, private-company data and market data, depending on which provider sits behind the connection. What arrives, and how, differs by provider:

Bigdata.com shows what a single provider app can bring into the conversation. Once connected, ChatGPT can search more than a billion financial documents, including filings, earnings transcripts, fundamentals and premium news, across an archive of more than 20 years, alongside real-time news from institutional sources and structured data for fundamentals, earnings and pricing. Its ten tools cover document search and full-text retrieval, security identification, tearsheets for companies, ETFs, sentiment, markets and countries, and a calendar of upcoming earnings, dividends and corporate events. Two credit-factor tools for screening and ranking companies by credit risk are listed as coming soon (Bigdata.com).

Two points are worth keeping in mind. First, “live” depends on the provider’s feed, not ChatGPT itself. If OpenAI indexes and hosts the data, its freshness depends on how often that index is updated. It is therefore worth asking the provider how quickly new filings and transcripts become available. Second, access to one dataset does not mean access to the provider’s full catalogue. For example, VentureBeat notes that the inclusion of LSEG News does not include every dataset offered by LSEG.

Coverage is useful, but citations are just as important. OpenAI says its hosted data can link figures to specific tables and passages. S&P Global also describes its retrieval as deterministic, meaning the same query returns the same cited result. If a connection cannot show where a claim came from, it becomes difficult to check or audit. Bigdata.com, for example, attaches a named source, date and URL to each claim, so every figure can be traced back to where it came from.

There are also some limits to how these connections work. According to OpenAI’s developer mode guide, OpenAI-built apps support search only. Deep research can use custom apps for reading and fetching information, while agent mode does not currently support custom apps. For most financial research, where the main task is finding and reading information rather than changing it, these limits may not be a major issue.

How to set it up, step by step

The plugin route takes six steps, and only the last one needs an admin. It follows OpenAI’s plugin guide.

1. Check your plan and workspace. The plugin directory is available across plans, but installing an individual plugin depends on your plan, workspace, role and region. In a managed workspace, ask an admin whether plugins are enabled for you.

2. Confirm your licence covers AI access. Ask the data provider in writing before you connect anything.

3. Find the plugin. Open Plugins in the ChatGPT sidebar, select the provider’s plugin, and read its included apps and setup requirements. For S&P Global, for example, you connect the Deterministic and Adaptive plugins.

4. Install and connect. Select Install plugin, then Connect on any app that needs authorisation, and sign in with the account your firm licensed.

5. Test with something you can check by hand. Ask for a figure you already know, then open its citation and confirm it matches the source.

6. Set the team policy. An admin opens Workspace settings, then Plugins, chooses Available or Installed, and reviews the allowed actions for each included app.

Worked example: Bigdata.com. Bigdata.com is listed as an official app rather than a custom MCP app, so there is no developer mode, server endpoint or tool scan to manage. In a managed workspace, an admin enables it once: open Workspace settings, go to Apps, search for Bigdata, select the Bigdata.com app and choose Enable. The setup guide then recommends opening the app’s Action consent setting and allowing read actions without user confirmation, so analysts are not asked to approve every lookup. Because the tools only read data, this sits comfortably with a read-only policy.

Each analyst then connects their own account. In Settings, go to Apps, choose Add more, select Bigdata.com, then Connect, and sign in with Bigdata.com credentials; the browser returns to ChatGPT once the login succeeds. To call the app, mention @Bigdata.com in a prompt, for example by asking for a company profile of Micron. The guide also suggests saving an instruction to ChatGPT’s memory telling it to use Bigdata tools instead of web search for financial, market and economic news, which stops it falling back to the open web for questions the licensed source can answer.

One catch for early adopters: if your team built a custom connector to the Bigdata.com MCP server before the official app existed, delete it and enable the official app instead. Both point to the same server, and ChatGPT does not allow two connectors with the same URL.

A custom MCP app follows a different path. An admin enables developer mode, goes to Apps and selects Create, enters the server endpoint and authentication method, runs Scan Tools, and saves the app as a draft. After testing, an admin publishes it. Two details catch people out. ChatGPT connects to remote MCP servers only, so a server on a private network needs OpenAI’s Secure MCP Tunnel. And if you use OAuth, your identity provider must issue refresh tokens, otherwise users have to sign in again once the first authorisation expires (OpenAI).

ChatGPT for Financial Services is not self-service. OpenAI makes it available to eligible financial institutions through its sales team, so the first step is a conversation, not a settings page.

What does the setup cost you in time and in licensing?

A provider plugin is usually quick to set up, while a custom connection takes more time. In either case, the main limitation is often the data licence. ChatGPT may support the connection, but the provider decides which data your account is allowed to use.

Time. An individual can usually install a provider’s plugin and sign in during one session. A custom MCP app requires a hosted server, authentication and an administrator to test and publish it. On Business, the app must be recreated when its tools change. OpenAI also keeps a fixed snapshot of the tools in an approved app, so later changes may cause errors until an administrator reviews them. If you have a custom GPT connected to a data API, allow time to rebuild it before 11 December because its actions will not migrate. At the quick end of the range, Bigdata.com puts its own ChatGPT setup at a few minutes.

Licensing. There are usually two costs to consider: the ChatGPT plan that supports the connection and the provider’s licence for the underlying data. The datasets included with ChatGPT for Financial Services come without separate contracts to negotiate. Access to subscription services such as FactSet or S&P Global still depends on the licence your firm holds with that provider. OpenAI does not publish pricing for the financial services plan and asks interested firms to contact its sales team. In Enterprise workspaces, actions in Work also use credits, so usage may carry an additional cost. Provider apps sit on top of these costs: Bigdata.com, for example, needs its own account, and offers a 14-day trial for teams that want to test it in ChatGPT before committing.

Compliance. OpenAI says business data in the financial services plan is not used for training by default and is encrypted both in transit and at rest. Firms can set retention periods, export supported workspace logs to their audit systems and create separate workspaces to maintain information barriers. Data providers may have their own logging policies, so these should be reviewed separately. Third-party MCP servers should also be checked before use, as an unsafe server can increase the risk of prompt injection.

What to check before you roll it out

Before giving access beyond the pilot group, check that the following points are covered:

• Each provider has confirmed in writing that your licence allows its data to be used with AI.

• You know how soon new filings and transcripts appear in each provider’s data.

• Every figure in a test response links to a citation that supports it.

• You understand what OpenAI and the provider record, where those logs are stored and how long they are kept.

• The compliance team has reviewed the setup, including any information barriers.

• App actions are kept read-only unless there is a clear need for write access.

• One person is responsible for the setup and reviews it whenever a tool definition changes.

• Any custom GPT that uses actions has a replacement plan in place before 11 December 2026.

• The same test questions are run again after every change to a plugin or app.