MCP Server

Overview

The Model Context Protocol (MCP) is an open standard that lets AI agents interact with external tools and services. The TestKase Test Management MCP Server (@testkase/test-management-mcp, formerly @testkase/mcp-server) connects any MCP-compatible AI agent to your TestKase workspace — giving it full access to create test cases, run cycles, pull reports, and manage your entire test lifecycle through natural language.

Supported AI agents:

• Claude Desktop — Anthropic's desktop app
• GitHub Copilot — VS Code agent mode
• Claude Code — Anthropic's CLI agent
• Cursor — AI-powered code editor

Prerequisites

Before setting up the MCP server, make sure you have:

1. A TestKase account — sign up free if you don't have one.
2. A Personal Access Token (PAT) — generate one from Settings → API Keys. See the API Keys guide for details.
3. An MCP-compatible AI agent — Claude Desktop, GitHub Copilot (VS Code), Claude Code, or Cursor.

Installation & Setup

The MCP server runs via npx — no global install needed. Configure your AI agent with the JSON config below.

Claude Desktop

Add to your Claude Desktop config file:

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

{
  "mcpServers": {
    "testkase": {
      "command": "npx",
      "args": ["-y", "@testkase/test-management-mcp"],
      "env": {
        "TESTKASE_API_BASE_URL": "https://api.testkase.com",
        "TESTKASE_PAT_TOKEN": "YOUR_PAT_TOKEN"
      }
    }
  }
}

1. Open or create the config file at the path above.
2. Paste the JSON block, replacing YOUR_PAT_TOKEN with your actual token.
3. Save the file and restart Claude Desktop.
4. Look for the hammer icon (🔨) in the chat — it shows available MCP tools.

GitHub Copilot (VS Code)

Add to your Copilot MCP config file:

• Windows: %APPDATA%\Code\User\globalStorage\github.copilot\mcp.json
• macOS: ~/Library/Application Support/Code/User/globalStorage/github.copilot/mcp.json

{
  "mcpServers": {
    "testkase": {
      "command": "npx",
      "args": ["-y", "@testkase/test-management-mcp"],
      "env": {
        "TESTKASE_API_BASE_URL": "https://api.testkase.com",
        "TESTKASE_PAT_TOKEN": "YOUR_PAT_TOKEN"
      }
    }
  }
}

1. Open or create the config file at the path above.
2. Paste the JSON block, replacing YOUR_PAT_TOKEN with your actual token.
3. Restart VS Code or reload the Copilot extension.

Claude Code (CLI)

Run this single command to register the MCP server:

claude mcp add testkase \
  -e TESTKASE_API_BASE_URL=https://api.testkase.com \
  -e TESTKASE_PAT_TOKEN=YOUR_PAT_TOKEN \
  -- npx -y @testkase/test-management-mcp

That's it — restart Claude Code and the tools are available immediately.

Cursor

Add to .cursor/mcp.json in your project root:

{
  "mcpServers": {
    "testkase": {
      "command": "npx",
      "args": ["-y", "@testkase/test-management-mcp"],
      "env": {
        "TESTKASE_API_BASE_URL": "https://api.testkase.com",
        "TESTKASE_PAT_TOKEN": "YOUR_PAT_TOKEN"
      }
    }
  }
}

1. Create the .cursor directory in your project root if it doesn't exist.
2. Create mcp.json with the config above, replacing YOUR_PAT_TOKEN.
3. Restart Cursor to pick up the new MCP server.

Available Tools

The MCP server exposes 11 tools covering the full test management lifecycle:

Field values are project-specific

Priority, status and the other dropdown fields are configured per project, so a value that works in one project is rejected in another. get_project_structure is the discovery call: ask it for field_options and it returns the values that project actually accepts, along with folder IDs, labels and members. The tool descriptions tell the agent to call it before creating or updating anything, and when a write fails on an invalid value the server returns a hint pointing back at it.

If your agent reports "Invalid priority value" or "folder_id must be a number", the fix is almost always to look up the real value rather than to guess a different one. Field options are cached for five minutes, so a change you make in Project Settings can take that long to be visible to the agent. Letter case doesn't matter: critical is accepted for Critical.

Small replies, fewer tokens

From version 2.5.0 the server keeps your AI assistant's context small:

• Lists return the fields an agent works with (ID, title, priority, status), plus one link pattern such as …/testcases/{id} instead of a full link on every row.
• Empty values and wrapper fields are left out.
• Add verbose: true to any read call when you need the full API response.

Compared with 2.4.1, the tool descriptions an assistant loads are about half the size, and replies are about two thirds smaller.

Works with smaller models

The tools are built so that smaller, cheaper models get them right too:

• Test steps, bulk test cases and results go in as plain JSON lists.
• Common spellings are understood: Passed, FAILED, not run, In Progress.
• When a call is wrong, the error says what to send, with an example.
• In a bulk create, a folder or priority given once applies to every test case.

Reporting

The get_report tool covers execution, coverage, trends, team, defect, and insight report types. Five of the 42 -- marked below -- have no card in the Reports tab, so the MCP server and the REST API are the only ways to read them:

✱ API-only — no card in the Reports tab, and not counted in the 37 reports the UI lists. See Reports Available Only Through the API for what each one returns.

Every call takes projectId and report_type. Most report types also accept one test_cycle_id, one test_plan_id, a folder_id, and startDate and endDate (YYYY-MM-DD, both days included); the time-series types accept granularity of day, week (the default) or month, and cycle_comparison takes cycle_ids. You do not need to remember any of that -- ask for a report the way you would ask a colleague ("show me the project health score for PRJ-1001") and the agent fills in the arguments. An unrecognised report_type comes back with the full list of valid types.

Configuration

Example Conversations

Here are some things you can say to your AI agent once the MCP server is connected:

Troubleshooting