ADX Brings Live Market Data to ChatGPT and Claude: What Users Actually Get
Abu Dhabi Securities Exchange has opened a new route into its market data. Investors and analysts can now query official ADX information from compatible AI applications, including ChatGPT and Claude, through a remote Model Context Protocol server.
The announcement arrived on 13 August. ADX calls it the first service of its kind from a MENA exchange. That claim is worth attributing to the exchange rather than treating it as independently proven, but the product itself is live: ADX has published individual plans, prices and usage limits.
This is more useful than another chatbot that knows yesterday's closing price. It gives an AI application a governed connection to exchange data. It does not, however, turn ChatGPT into a trading terminal or remove the need to check the answer.
What ADX has launched
The service places a remote MCP server between ADX data and a compatible AI application. MCP is a standard that lets an AI client discover and call external tools. In this case, a user can ask a natural-language question, the client calls the ADX tool, and the returned data is used to compose the answer.
According to the official ADX plan page, the available information covers market depth and bid/ask data, trading activity split by retail or institutional investors and by foreign or local investors, corporate actions, announcements, XBRL financial disclosures, index constituents and reference data. The TRADE also reported that the connection is intended to remove some of the terminal and API work normally needed to query exchange data.
That does not mean every response is simply generated from a fresh market-data call. Users still need an ADX account and entitlement, a compatible AI product, the remote connector configuration and enough quota for the request. The AI client also decides when to call the tool and how to explain the result.
Four plans, with a low-cost entry point
ADX lists four individual plans. Starter is free and includes 100 tool calls per month with current or last trading-day history. Essential costs AED 9.99 per month for 500 calls and two years of history. Advanced costs AED 29.99 for 1,500 calls and five years. Premium costs AED 49.99 for 4,000 calls and full history.
All four are listed as including real-time Level 1 data for all instruments. The practical differences are quota and historical depth, not whether the cheapest plan receives a delayed Level 1 feed. Gulf News and the WAM announcement record independently confirm the free-to-AED-49.99 price range, although both reports originate from the same launch announcement.
ADX individual-plan prices, quotas and history depth, plus the connection path and verification checks. Sources: ADX MCP plans; OpenAI and Anthropic connector documentation. Checked 17 August 2026. Credit: SultanByte editorial artwork.
For occasional checks, 100 calls may be enough. A user who repeatedly refines a question, compares several instruments or asks the client to retrieve multiple datasets could consume calls much faster than expected. ADX's public plan page describes monthly tool-call limits, not a fixed number of complete conversations.
ChatGPT and Claude access is not identical
The exchange headline says the data is available through ChatGPT and Claude. The platform rules matter.
OpenAI's developer mode and MCP documentation says full MCP support is in beta for Business and Enterprise/Edu workspaces on the web. Pro users can connect remote servers that expose read or fetch actions through developer mode. In managed workspaces, an administrator may need to review and publish the custom app before other users can access it.
Anthropic's remote MCP connector documentation lists support across Free, Pro, Max, Team and Enterprise. Free users are limited to one custom connector, while Team and Enterprise setup requires an owner.
These conditions can change quickly. A procurement team should verify the supported plan, admin controls and connector permissions in the vendor documentation before buying ADX capacity for a larger group.
Better provenance, but not an automatic green light
Official data access solves one common problem with general-purpose AI: the model no longer has to guess a price from old training data or an unverified web page. The answer can be grounded in a tool response from the exchange.
The interpretation can still be wrong. A model may choose the wrong instrument, mix a latest trade with a bid or ask, overlook the market status, misunderstand the requested period or summarize a returned table poorly. A plausible paragraph is not evidence that the correct tool ran with the correct parameters.
Before using an answer in research or a client workflow, check four things:
Confirm the instrument and exchange code.
Check the timestamp, market status and requested period.
Inspect the returned source data, not only the prose summary.
Keep order execution outside the conversational workflow unless a separately controlled and documented trading system handles it.
Both OpenAI and Anthropic warn users to trust remote MCP servers carefully because external connectors can introduce prompt-injection and data-sharing risks. ADX describes its server as governed and secure, but the reviewed public material does not include an independent security audit, uptime history, latency benchmark or accuracy test. Firms should treat those as vendor-assessment questions rather than assume the connector has inherited the controls of an institutional terminal.
Why this matters beyond one UAE exchange
The useful regional precedent is the delivery model. Gulf exchanges have spent years improving data distribution, investor access and digital services. MCP gives them another channel: official data can reach the AI interface that an analyst already uses, without forcing the analyst to build a bespoke integration first.
That approach also changes where product boundaries sit. SultanByte's comparison of UAE Open Finance and Saudi Open Banking showed why permissioning, data scope and liability matter more than a polished front end. The same test applies here. The conversational interface is the visible part; entitlements, quotas, source provenance and audit logs decide whether the service is suitable for production.
Other exchanges may follow with their own connectors, but the region should not be treated as a single market. Each venue has different licensing rules, data products, instrument coverage and technology priorities. An MCP endpoint from one exchange does not create a common GCC market-data layer.
A useful interface, with clear limits
ADX has made official market information easier to query and priced the entry tier low enough for individual experimentation. The strongest use cases are research assistance, company screening, disclosure retrieval and quick market checks where the user can inspect the underlying result.
The service is not a substitute for execution controls, independent validation or a licensed professional terminal when those are required. Teams evaluating it should run a short test set across live and historical questions, record tool usage, compare every answer with the source output and review the connector's security model. If it passes those checks, conversational access can remove real friction without pretending that a fluent answer is the same thing as a verified one.
