QTM4J MCP Server

An MCP server that exposes the QMetry Test Management for Jira Cloud (QTM4J) REST API as tools Claude (or any MCP-compatible client) can call.

Features

Tools cover the most common CRUD flows across the major QTM4J entities:

All tools validate inputs with Zod, paginate list endpoints via startAt / maxResults, and automatically retry rate-limited (HTTP 429) responses with exponential back-off up to 3 attempts.

Requirements

• Node.js 18+ (uses native fetch)
• A QMetry API key (from QMetry → API Keys)

Setup (same steps for all clients)

1. Clone, install, build:

git clone https://github.com/salehrifai42/qmetrymcp.git
cd qmetrymcp
npm install
npm run build

2. Add your API key — copy the example and fill in your key:

cp .env.example .env

Then edit .env:

QTM4J_API_KEY=your-api-key-here
QTM4J_REGION=US

.env is git-ignored and never committed.

3. Open your editor/IDE in the repo directory. The server loads .env automatically on startup — no extra config needed:

• Claude Code — run claude in this directory. .mcp.json is pre-configured and the qtm4j server connects automatically.
• GitHub Copilot (VS Code) — open the folder in VS Code, switch Copilot Chat to Agent mode. .vscode/mcp.json is pre-configured and the qtm4j tools appear automatically.

Configuration

The server reads two environment variables:

Local config reference

A config.template.json is included with common field values (project ID, statuses, priorities, execution result IDs, folder IDs, etc.). Copy it for local reference:

cp config.template.json config.json

config.json is git-ignored. The MCP server reads only from environment variables.

Running manually

QTM4J_API_KEY=your-key npm start

The server speaks MCP over stdio — you don't normally run it directly; your MCP client spawns it.

MCP client configuration

Claude Code (CLI)

.mcp.json ships with this repo and is auto-detected when you open Claude Code in the project directory. Verify:

claude mcp list

Run /mcp inside a session to see connected servers and their tools.

Claude Desktop

Edit ~/Library/Application Support/Claude/claude_desktop_config.json on macOS and restart Claude Desktop:

{
  "mcpServers": {
    "qtm4j": {
      "command": "node",
      "args": ["/path/to/qmetrymcp/dist/index.js"],
      "env": {
        "QTM4J_API_KEY": "your-api-key-here",
        "QTM4J_REGION": "US"
      }
    }
  }
}

GitHub Copilot (VS Code)

.vscode/mcp.json ships with this repo and is auto-detected when you open the folder in VS Code. Switch Copilot Chat to Agent mode and the qtm4j tools appear automatically.

Trying it out

Once connected, ask the assistant something like:

Search QMetry project 10011 for test cases with status "To Do" and show me the first 5.

The client will call search_test_cases with { projectId: 10011, status: ["To Do"], maxResults: 5 } and render the response.

Get all executions in test cycle FS-TR-747 and mark any unexecuted ones as Pass.

Example tool calls

// Search test cases in project with numeric ID 10011
{
  "name": "search_test_cases",
  "arguments": {
    "projectId": 10011,
    "status": ["Approved"],
    "maxResults": 20
  }
}

// Update an execution result (executionResultId: 239444=Pass, 239441=Fail, 239443=Not Executed)
{
  "name": "update_test_execution",
  "arguments": {
    "cycleId": "gxMbioKJsyEr3E",
    "testCaseExecutionId": 287595809,
    "executionResultId": 239444,
    "comment": "Verified on staging"
  }
}

Error handling

• Non-2xx responses return a tool error with the HTTP status and parsed API body.
• Network errors return a descriptive error message.
• 429 responses are retried automatically with exponential back-off (up to 3 total attempts).

Notes

• projectId must be the numeric Jira project ID (e.g. 10011), not the project key (e.g. "FS"). You can find it in the Jira project URL: …?projectId=10011&projectKey=FS.
• Search endpoints use POST /…/search — filters go in the body under filter, pagination/sort on the query string. The MCP handlers wrap this for you automatically.
• "Update" endpoints that return 204 No Content resolve with a simple { message: "…" } payload.
• The Swagger spec does not currently document a framework-style automation import-result endpoint (e.g. JUnit/TestNG/Cucumber ingestion); the automation tools here cover the rules-run and rule-link flows exposed in the spec.