MCP Workflow Examples

Start with a task that can be checked by reading: project structure, data sources, parameters and strategy decisions. Ask for file and method references rather than a generic description of an assumed algorithm.

A useful answer names the actual project and identifies the code responsible for entries, exits and risk. Missing information should be reported as a missing file or result.

Inspect the selected XTester project read-only. Explain entries, exits, parameters and risk controls, citing a file and method for each finding. Separately list what cannot be established without running it.

Editing requires suitable access and writer ownership. Specify one concrete change so the agent can return a small, checkable result.

Check the installed API first, apply the edit through project tools, then compile. Successful compilation establishes build validity; strategy behaviour requires a separate run.

Add a maximum-position-size parameter to the selected XTester strategy using the installed API. Show the file changes and compilation result, then list the behavioural checks still needed.

Specify the strategy, exchange, market, instrument, timeframe, period and starting funds. Review costs and data coverage before running. Insufficient data should produce an explanation, not substitution with another history.

Retain the run ID, conditions and result sources. Use runId to revisit metrics, trades and equity. Distinguish an executed calculation from a proposed plan.

Prepare a backtest of the selected XTester strategy for the instrument and period I specify. Check history, starting funds and costs first. Return the actual runId, conditions, metrics and trades, or the precise reason the run could not execute.

Compare versions with the same data, period, capital and costs. Identify changed parameters and the baseline; otherwise differences may come from run conditions.

Ask for trades, drawdown and behaviour across different segments as well as final profit. Set the selection criterion before comparing.

Compare these two XTester run IDs. Verify data and costs, list differing conditions, then compare trades and drawdown against the stated criteria and explain the limits of the conclusion.

In Settings → External Agents, Access and visibility shows the mode for new connections and whether agent work is shown. Confirmation settings opens Studio’s corresponding preferences.

Application screenshotXTester 1.0.0-preview.1

Enlarge screenshot

xtester.mcp.autonomousAgents allows trusted agents to connect and perform ordinary project work without separate Studio prompts. Account access and protected operations retain their own rules. An agent cannot enable this setting itself.

xtester.mcp.followAgentWork controls opening the relevant views and files while an agent works. It grants no permissions. With following disabled, authorised work continues while your current view stays unchanged.

Inspect and revoke remembered grants in the same section. A read-only agent needs a new suitable access request to edit; its name or the visibility setting cannot increase its role.

No server appears in the client. Check the settings file, xtester-studio name, JSON or TOML syntax and absolute path. Refresh the server or restart the client after editing.

The MCP process does not start. Open Studio and copy configuration again; a move or update may invalidate the old path. If Studio reports a missing component, check the application installation.

SELECTION_REQUIRED. The client has not selected a live window or lost its previous selection. List sessions again and choose the exact window.

APPROVAL_REQUIRED or pending. A decision is awaited in Studio. Check the intended window; rejection or expiry does not grant access.

SESSION_GONE. The window or session closed or changed. Obtain a fresh list; identities from a previous run are stale.

Reading works but editing fails. Check access and writer status. Changes require suitable approval and available or explicitly transferred writer ownership.

The application is busy. Inspect the reason and current job; wait or use supported cancellation for your own task rather than launching duplicates.

Include the XTester version, OS, AI client and failing step: registration, process startup, window selection, approval or tool call. Supply the exact error and the project actually selected.

For configuration, the server name, command and arguments with personal paths anonymised are sufficient. Exclude passwords, tokens and exchange secrets.