QTM4J MCP Server

An MCP server with 87 tools for QMetry Test Management for Jira (QTM4J). Search and manage test cases, cycles, executions, plans, folders, comments, defects, automation rules, and project metadata from Claude Desktop, Claude Code, VS Code Copilot, Cursor, or any MCP-compatible client.

Distribution:

• npm: qtm4j-mcp-server
• MCP Registry: io.github.salehrifai42/qtm4j-mcp-server
• GitHub: salehrifai42/qmetrymcp

Quick start (no clone required)

You need a QMetry API key (QMetry → API Keys) and Node.js 18+.

Claude Desktop

Edit your config file and restart Claude:

• macOS: ~/Library/Application Support/Claude/claude_desktop_config.json
• Windows: %APPDATA%\Claude\claude_desktop_config.json
• Linux: ~/.config/Claude/claude_desktop_config.json

{
  "mcpServers": {
    "qtm4j": {
      "command": "npx",
      "args": ["-y", "qtm4j-mcp-server@^0.1"],
      "env": {
        "QTM4J_API_KEY": "your-api-key-here",
        "QTM4J_REGION": "US"
      }
    }
  }
}

Claude Code (CLI)

claude mcp add qtm4j -e QTM4J_API_KEY=your-api-key-here -e QTM4J_REGION=US -- npx -y qtm4j-mcp-server@^0.1

VS Code (GitHub Copilot Agent mode)

Create .vscode/mcp.json in your workspace:

{
  "servers": {
    "qtm4j": {
      "type": "stdio",
      "command": "npx",
      "args": ["-y", "qtm4j-mcp-server@^0.1"],
      "env": {
        "QTM4J_API_KEY": "${env:QTM4J_API_KEY}",
        "QTM4J_REGION": "US"
      }
    }
  }
}

Switch Copilot Chat to Agent mode and the qtm4j_* tools appear automatically.

Cursor

Add to ~/.cursor/mcp.json (or <project>/.cursor/mcp.json for project-level):

{
  "mcpServers": {
    "qtm4j": {
      "command": "npx",
      "args": ["-y", "qtm4j-mcp-server@^0.1"],
      "env": { "QTM4J_API_KEY": "your-api-key-here" }
    }
  }
}

The example configs above pin the package to ^0.1 so a future breaking release won't auto-upgrade you. Drop the @^0.1 suffix if you'd rather always run the latest.

Global install (faster startup)

npx re-resolves the package on every launch, which adds a few seconds of startup latency. If you use the server frequently, install it globally and point your client at the binary directly:

npm install -g qtm4j-mcp-server

Then in your client config, replace the npx command:

{
  "mcpServers": {
    "qtm4j": {
      "command": "qtm4j-mcp-server",
      "env": {
        "QTM4J_API_KEY": "your-api-key-here",
        "QTM4J_REGION": "US"
      }
    }
  }
}

Tradeoff: instant startup and works offline, but you'll need to run npm update -g qtm4j-mcp-server to get new versions.

Configuration

💡 Set QTM4J_REGION=AU if your QMetry instance is on the Sydney cluster.

Tools

All tools are prefixed with qtm4j_ to avoid collisions with other MCP servers.

See docs/TOOLS.md for the full reference and docs/COOKBOOK.md for example prompts colleagues can paste into any MCP client.

Every tool ships with annotations (readOnlyHint, destructiveHint, idempotentHint, openWorldHint) so clients can decide whether to ask for confirmation. Read tools accept a response_format parameter (json default, or markdown for human-readable output). Large responses are automatically truncated at 25k characters with a hint to narrow the query.

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

Trying it out

Once connected, ask your assistant something like:

Search QMetry project <your project ID> for test cases with status "To Do" and show me the first 5.

The client will call qtm4j_search_test_cases and render the response.

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

Example tool calls

Replace <projectId> and <KEY> with your tenant's numeric Jira project ID and project key. Find them in the project URL: …?projectId=<projectId>&projectKey=<KEY>. Use qtm4j_get_execution_results to discover your tenant's executionResultId values.

// Search test cases
{
  "name": "qtm4j_search_test_cases",
  "arguments": {
    "projectId": "<projectId>",
    "status": ["Approved"],
    "maxResults": 20,
    "response_format": "markdown"
  }
}

// Update an execution result
{
  "name": "qtm4j_update_test_execution",
  "arguments": {
    "cycleId": "<internal-cycle-id>",
    "testCaseExecutionId": "<internal-tc-execution-id>",
    "executionResultId": "<pass-id>",
    "comment": "Verified on staging"
  }
}

Troubleshooting

• Tools don't appear in my client. Restart the client after editing config. Check claude mcp list (Claude Code) or VS Code's MCP panel for connection status. On first run, npx -y qtm4j-mcp-server may take a few seconds to download the package.
• 401 Unauthorized. Your QTM4J_API_KEY is invalid or expired. Generate a new one in QMetry → API Keys.
• 404 on execution or search endpoints. Many endpoints want the internal numeric id, not the human key like <KEY>-TR-747. Call qtm4j_get_test_cycle first to translate the key into the internal id.
• Empty or oversized folder response. Pass folderId to qtm4j_list_folders to scope to a subtree — full project trees on large projects can exceed the response size limit.
• projectId rejected. Use the numeric Jira project ID, not the project key string. You can find it in the Jira project URL: …?projectId=<numeric-id>&projectKey=<KEY>.

Notes

• Search endpoints use POST /…/search — filters go in the body under filter, pagination/sort on the query string. Tool handlers wrap this for you.
• 204 No Content responses resolve as { message: "…" }.
• The Swagger spec does not currently document a framework-style automation import-result endpoint (e.g. JUnit/TestNG/Cucumber ingestion); the automation tools cover the rules-run and rule-link flows exposed in the spec.

Development

Local setup if you want to modify the server:

git clone https://github.com/salehrifai42/qmetrymcp.git
cd qmetrymcp
npm install
npm run build
QTM4J_API_KEY=your-key npm start

Test changes with the MCP Inspector:

QTM4J_API_KEY=your-key npx @modelcontextprotocol/inspector node dist/index.js

Run the unit tests (no API key or network needed — the HTTP client's transport is faked in-memory):

npm test

Replicating the bulk xlsx import workflow