XTester MCP Gives AI Agents a Strategy-Testing Environment
XTester MCP connects an AI agent to a local environment for developing and historically testing strategies, including the project, compiler, market data and test results.
AI can already turn a trading idea into code. The harder question comes next: who verifies that the code compiles, the test uses the intended history, and an improved metric did not appear because fees changed or the execution model became more forgiving?
XTester MCP connects an AI agent to a local environment for developing and historically testing strategies. This changes the workflow: the agent works not only with text in a chat, but with a project, compiler, market data and test results.
From a chat response to an inspectable loop
Without access to a testing environment, the usual process is fragmented:
- the model suggests C# code;
- the user transfers it into an application;
- compilation errors are fixed manually;
- a test is launched;
- part of the result is copied back into the chat;
- the model proposes another version without seeing the full context.
Every transfer creates risk. Settings can be lost, different date ranges can be compared, or only the most attractive part of the output can reach the model.
Through MCP, an agent can follow a connected research loop: inspect the environment, modify the strategy inside an XTester project, compile it, check history, run a backtest and inspect trades, metrics, the equity curve and drawdown.
Example: one hypothesis, one change
Suppose the goal is to test whether a volatility filter improves a breakout strategy.
A controlled sequence would be:
- freeze the baseline strategy version;
- select the venue, instrument and historical range;
- preserve fees, funding, slippage and other execution assumptions;
- run the baseline;
- change only the filter;
- repeat the test under the same conditions;
- compare trades, drawdown and stability rather than only the final number.
AI contributes speed when working with code and outputs. Experimental control remains mandatory.
Why XTester remains a local application
XTester MCP uses stdio: the client starts the host on the user’s computer, and XTester does not open a separate network MCP port. Setup happens through the built-in XTester screen, which knows the actual path of the installed release.
“Local MCP” does not mean that the complete chain is offline. Claude, Codex or another client may call a cloud model and transmit data according to its configuration. Market-history downloads, GitHub operations and updates can also use the network.
Before connecting an agent, users should therefore inspect the AI client as well as XTester: retention, telemetry, model provider and how much data is included in tool results.
What is available now
XTester now has a dedicated MCP documentation section covering:
- the architecture and limitations;
- setup through the built-in connection screen;
- capability classes;
- conceptual scenarios;
- security and data boundaries;
- machine-readable discovery for search engines and AI crawlers.
Exact tool names, arguments and schemas must be published from the matching verified release catalog. Until that export is complete, the documentation intentionally avoids inventing commands or an exact tool count.
Why this matters to EasyTrading
A strategy passes through several stages across the EasyTrading ecosystem. XTester handles development and historical testing. Other components solve separate execution and analysis problems.
MCP improves the research stage specifically. An AI agent can move faster from a hypothesis to an inspectable result, while the user can still see the code, run conditions and limitations of the conclusion.
This does not turn research into a shortcut to deployment. It adds a more formal testing layer between an idea and any execution decision.
The connection, project and results workflow is explained in detail in XTester MCP: How an AI Agent Works with a Local Backtester.
Technical details and setup are available in the XTester documentation:
XTester is designed for strategy development and historical testing. A backtest does not guarantee future returns and is not investment advice.