Skip to main content
Version: 3.2

Agentic Testing Overview

This guide describes the system architecture and unified interaction workflows for running Agentic Test Automation with Menditect Test Automation (MTA) and AI assistants (such as Cursor, Claude Code, GitHub Copilot, Gemini / Antigravity, and Cline).

1. System Overview​

Agentic Test Automation connects four decoupled systems that communicate through standard protocols (MCP, HTTPS, and WebSockets):

The 4 Core Systems​

SystemRolePrimary Interface / Protocol
1. MTA Platform (MTA Server)Central test management repository, orchestration engine, and authoring backend. Hosts test suites, test cases, and historical runs. Exposes MTA Primitive Tools MCP for authoring and platform dispatch.HTTPS / SSE MCP (/primitivetools/mcp)
2. MTA Mendix Plugin (MtaPluginModule)Embedded module running inside your Mendix application. Connects via WebSocket to MTA for formal test dispatch, and exposes an embedded HTTP MCP server (/plugin/mcp) for in-memory exploratory test execution (execute-testcase).WebSocket (MTA Connection) & HTTP/SSE MCP (/plugin/mcp)
3. Mendix Model Discovery (mxcli / Studio Pro MCP)AST inspection engines parsing domain entities, microflows, nanoflows, pages, and widgets. Provides semantic app context to the AI assistant.CLI Stdio (./mxcli) or HTTP MCP (localhost:7782/mcp)
4. Menditect Agentic Test Skills (agentic-test-skills)Structured testing intelligence guiding AI assistants. Enforces conversational state routing (AGENTS.md), 9-section Execution Plans (EP_*.md), layered step construction, and failure diagnostics.Markdown / Skill Standard (skills/ or skillssource/_modules)

2. Unified System Interaction Workflow​

Whether executing a rapid, in-memory exploratory test or authoring a permanent test suite in the MTA repository, the lifecycle follows a single unified workflow.

Both pathways share the same discovery, model analysis, and execution plan design phases. The only difference is the execution target: the local runtime MCP plugin (for sub-second in-memory tests with automatic rollback) or the MTA Platform MCP (for permanent suite authoring and platform test dispatch).

3. Configuration Contract: mta_config.json​

The mta_config.json document serves as the Single Source of Truth (SSOT) for AI assistants, providing all endpoints, directory paths, application instances, and model discovery sources.

Canonical Structure​

{
  "$schema": "./mta_config.schema.json",
  "workspace_type": "clone_root",
  "workspace_dir": "C:\\Projecten\\my-workspace",
  "skills_dir": "C:\\Projecten\\my-workspace\\skills",
  "skills_style": "standard",
  "mta_output_path": "C:\\Projecten\\my-workspace\\menditect-output",
  "execution_plans_dir": "C:\\Projecten\\my-workspace\\menditect-output\\execution-plans",
  "mendix_version": "11.12.011",
  "application_name": "CarRental_App",
  "mta_base_url": "https://mta-trial.mendixcloud.com",
  "mcp_endpoint": "https://mta-trial.mendixcloud.com/primitivetools/mcp",
  "plugin_mcp_url": "http://localhost:8081/plugin/mcp",
  "playwright_viewer_url": "https://trace.playwright.dev/?trace=",
  "tracefile_base_url": "http://localhost:8081/rest/private/tracefile?fileUUID=",
  "app_instances": [
    {
      "name": "Local Development",
      "token": "f5d3f2a8-eb39-4cf5-9dfc-7fdeaf79c80d",
      "mtaUrl": "https://mta-trial.mendixcloud.com",
      "runtimeUrl": "http://localhost:8081/",
      "pluginUrl": "http://localhost:8081/plugin/mcp",
      "pluginToken": "Bearer 1",
      "pluginPort": "8081"
    }
  ],
  "default_app_instance": "Local Development",
  "default_app_instance_token": "f5d3f2a8-eb39-4cf5-9dfc-7fdeaf79c80d",
  "model_source": "mxcli",
  "mendix_project_dir": "C:\\MendixProjects\\CarRental",
  "mendix_mpr_path": "C:\\MendixProjects\\CarRental\\CarRental.mpr"
}

Property Reference​

PropertyTypeRequiredDescription
application_namestringYesTarget application name in MTA. Eliminates manual application selection prompts.
mta_base_urlstring (URI)YesWeb portal URL of the MTA platform (e.g. https://mta-trial.mendixcloud.com).
mcp_endpointstring (URI)YesMTA Primitive Tools MCP endpoint ([mta_base_url]/primitivetools/mcp).
execution_plans_dirstringYesFilesystem path where active Execution Plans (EP_*.md) are persisted.
mendix_project_dirstringYesAbsolute directory path to the target Mendix project folder.
mendix_mpr_pathstringNoAbsolute path to the .mpr project file used by mxcli.
plugin_mcp_urlstring (URI)NoEmbedded MTA Plugin MCP URL (http://localhost:8081/plugin/mcp).
app_instancesarrayNoArray of environment profiles containing instance names, tokens, URLs, and ports.
default_app_instancestringNoName of the primary default environment (e.g. "Local Development").
default_app_instance_tokenstring (UUID)NoMTA Application Instance Token used automatically by ExecuteTest.
model_sourcestringNoModel discovery engine: "mxcli" (offline .mpr inspection) or "studiopro" (Studio Pro live MCP).
playwright_viewer_urlstring (URI)NoBase URL for the trace viewer interface (https://trace.playwright.dev/?trace=).
tracefile_base_urlstringNoEndpoint to download trace files ([RuntimeUrl]/rest/private/tracefile?fileUUID=).

Sensitive Tokens in .env​

Sensitive authentication tokens are maintained in .env rather than committed into mta_config.json:

# MTA Primitive Tools MCP Header
MTA_MCP_AUTH_HEADER="Bearer eyJhbGciOi..."

# MTA Plugin MCP Header
PLUGIN_MCP_TOKEN="Bearer 1"

# Application Instance Token (Fallback)
MTA_APPLICATION_INSTANCE_TOKEN="f5d3f2a8-eb39-4cf5-9dfc-7fdeaf79c80d"

Configuration Resolution Order​

AI assistants evaluate configuration sources in this strict order:

  1. mta_config.json (Priority #1 - SSOT)
  2. Project AGENTS.md Directives
  3. Environment Variables (.env)
  4. IDE Settings (settings.json)
  5. Interactive User Prompt (only if a required setting is absent across all sources)
Last updated on