MetaTrader 5 Now Ships Its Own MCP Server: What That Changes

If you have read our earlier guides on connecting AI assistants to MetaTrader 5, you know the drill: install a third-party MCP server, edit a JSON config, restart your client, and hope the bridge holds. That approach still works. But as of 23 July 2026 it is no longer the only way. And for most people it is no longer the best place to start.

MetaTrader 5 Build 6060 added native support for the Model Context Protocol and an integrated AI Assistant. MetaQuotes put it plainly: "With built-in MCP support, MetaTrader 5 becomes a fully featured source of market data and trading functionality for modern AI agents."

This article covers what actually shipped, how to switch it on, what it can and cannot do, and when the third-party route is still the right call.

What actually shipped

Three separate things arrived together, and it is worth keeping them distinct because they solve different problems.

1. Native MCP support in the terminal. MT5 itself now speaks MCP. AI agents can read market information, analyse charts, inspect the trading environment and perform operations through the standard interface, no bridge process sitting between them.

2. A built-in AI Assistant. A conversational panel inside the terminal, and a second one inside MetaEditor. This is the part most traders will actually touch.

3. External agent connectivity. If you would rather use your own agent, you can connect external MCP-capable systems, MetaQuotes names OpenAI Codex and Claude Code as examples.

Turning it on

The setup is close to nothing, which is the point.

Sign in to the platform with your MQL5.community account. Every user automatically receives the free MQL5 Lite plan. After updating, the platform configures the rest itself: MQL5.community is selected as the default AI provider and the API key is added automatically.

If you would rather not route through MQL5, you can supply your own keys and pick a model directly in the terminal settings. Supported providers include OpenAI, Anthropic, Gemini, DeepSeek and Ollama, the last of which means fully local inference if you want nothing leaving your machine.

AI settings synchronise between the terminal and MetaEditor, so you configure it once.

What the assistant can do

In the terminal, it works on the market and your account:

• Explain current conditions for a symbol, trend, support and resistance, what is driving the move.
• Review your open positions and flag which carry the most risk, with reasoning.
• Analyse trading history over a period and identify your most profitable patterns.
• Answer questions about instruments and recent market events.
• Scan Market Watch for breakouts or developing trends across symbols.

In MetaEditor, it is a development assistant:

• Generate new MQL5 programs from a natural-language description.
• Analyse existing code, detect errors and propose fixes.
• Explain complex algorithms and suggest refactors.
• Convert MQL4 Expert Advisors to MQL5.

Conversation history is kept in the Chats tab of the Navigator, so a session survives a restart.

The security controls are the important part

This is where the native implementation deserves credit, and where anyone building their own bridge should pay attention.

AI operations are governed by configurable settings. You can explicitly allow or prohibit AI-initiated trading operations, or require manual confirmation before any order is placed. Separate settings control whether the AI may make network requests or run command-line operations.

If you have read our beginner's guide, you will recognise the warning we made there: the same interface that lets you ask for candles will happily place an order. In the third-party servers, that risk is yours to manage in code. MetaQuotes has made it a first-class setting.

Build 6060 also added Passkey authentication, which protects against credential theft, even with your login and password, an attacker cannot connect without confirmation from your registered device or a hardware key. It is enabled broker-side, so not every account will have it yet.

Does this make the third-party servers redundant?

No, but it changes why you would use one.

The native implementation is aimed at a human sitting in front of the terminal, working conversationally. The third-party servers do something different: they expose MT5 as programmatic infrastructure.

Reach for a custom or third-party server when you need:

• Control from your own code, wiring MT5 into a Python pipeline, a scheduler, or a service that runs headless with no terminal UI involved.
• Custom guardrails, lot-size ceilings, daily loss caps, instrument allow-lists, or a confirmation step you designed rather than one the platform shipped.
• Shaping your own tool surface, exposing a deliberately narrow set of operations instead of everything the platform can do, which is a meaningful safety property when an autonomous agent is on the other end.
• Non-MCP consumers, the third-party servers can also run as a REST API or a WebSocket stream, which is useful when the thing on the other side does not speak MCP at all.

Put simply: the native MCP server is a better assistant. A custom server is still a better integration.

A decision guide

• You want to ask questions about your account and chart, and have AI help you write MQL5 → use the built-in assistant. Nothing to install.
• You want an agent to operate your terminal inside your own pipeline → build or install a server, and read our advanced guide.
• You want risk limits the platform does not offer → custom server. This is the strongest remaining reason to build your own.
• You are on macOS or Linux and want programmatic access → unchanged by this release. The MetaTrader5 Python package remains Windows-only.

What it does not solve

Worth being clear about, because the announcement is easy to over-read.

• It is not a strategy. An assistant that can read your positions and explain them is not an edge. The hard part, deciding what to trade, when, and how much, is untouched.
• It does not backtest for you. The Strategy Tester is not suddenly an agentic research engine. Automating that still means writing code.
• It does not remove the need to test on demo. If anything, an assistant that can place orders makes that advice more urgent, not less.
• It does not make models reliable. A model that confidently misreads a chart is still a model that confidently misreads a chart. The interface changed; the capability did not.

Where to go next

If you are new to this area, start with the built-in assistant, it costs you nothing but an update and a sign-in, and it will tell you quickly whether conversational access to your terminal is useful to you.

If you already built a bridge, keep it. The reason to have one has narrowed, but it has not disappeared: programmatic control, custom risk limits, and headless operation are all still things the native implementation does not give you.

And if you have not yet read our earlier pieces, they remain accurate on the mechanics of third-party MCP servers, just no longer the only path in:

• Connecting AI Agents to MetaTrader 5 with MCP: A Beginner's Guide
• Building a Custom MCP Server for MetaTrader 5

Details in this article are drawn from the MetaQuotes release announcement for MetaTrader 5 Build 6060, published 23 July 2026. Provider and feature availability may change, check the terminal's own settings for the current list.