MCP Server: Generate Manual Test Cases

This MCP server provides the generate_testcases tool to generate manual test cases from documentation and rules you supply.

Installation

npm install
npm run build

Running the server

• Production: npm start (runs node dist/index.js)

• Development: npm run dev (runs with ts-node)

The server communicates over stdio (stdin/stdout) and is intended to be started by Cursor or another MCP client.

Configuring in Cursor

Add to your MCP config (e.g. Cursor: Settings → MCP or ~/.cursor/mcp.json):

{
  "mcpServers": {
    "manual-testcases": {
      "command": "node",
      "args": ["/PATH/TO/PROJECT/dist/index.js"]
    }
  }
}

Example with a real path:

{
  "mcpServers": {
    "manual-testcases": {
      "command": "node",
      "args": ["/Users/huenguyen/Desktop/hue-data/hue-data/workspace/mcp-manual-testcases/dist/index.js"]
    }
  }
}

After adding this, Cursor will expose the generate_testcases tool from this server.

Tool: generate_testcases

Generates test cases from documentation and rules.

Parameters

Note: At least one of document_content or document_path is required.

How it works

1. Read document: From document_content or by reading the file at document_path. PDF files are supported (text is extracted automatically).

2. Combine with rules: Build a prompt from document + rules.

3. Generate test cases:

   • If use_llm === true and the client (e.g. Cursor) supports sampling (LLM): the server sends a request to the client so the LLM generates test cases and returns the result.

   • Otherwise (client does not support sampling or use_llm === false): the server returns the formatted prompt for you to copy and use with an external LLM.

Example rules

• "Output test cases as a table: ID, Description, Preconditions, Steps, Expected result, Priority."

• "Priority P1 for login/payment flows; P2 for secondary screens."

• "Output language: English."

• "Each scenario has at most 10 steps; split into multiple scenarios if more complex."

Example usage in Cursor

You can ask the AI in Cursor, for example:

• "Use the generate_testcases tool: document_path is docs/feature-login.md, rules are 'Table format, English, P1 for happy path'."

• Or paste document content and use document_content with rules.

Generate test cases locally (script)

The script reads doc (txt, md, or PDF) and rules, then either calls OpenAI to generate test cases or prints the formatted prompt.

Environment variables:

Examples:

# Print formatted prompt (no API key needed)
npm run generate-testcases

# Read requirement from PDF and output CSV (requires OPENAI_API_KEY)
export OPENAI_API_KEY=sk-your-key
export DOC_PATH=path/to/requirements.pdf
export RULES_PATH=samples/rules-example.txt
export OUTPUT_FORMAT=csv
export OUTPUT_FILE=samples/generated-testcases.csv
npm run generate-testcases

CSV columns match the rules format: 模块, 标题, 前置条件, 步骤描述, 预期结果, test1测试人员, test1测试结果, buglink, PRE测试人员, PRE测试结果, buglink.

Quick try with samples

The samples/ folder contains:

• doc-example.md — sample requirement doc (中控后台PC, 6.1–6.4). It references prototype images in Images/ (image1.png–image4.png).

• images/ (or Images/) — prototype screenshots; when present next to the doc, the script and MCP prompt include the image list so generated test cases can reference them (e.g. “参照原型 Images/image1.png”).

• rules-example.txt — sample rules (table columns, 模块 format 代理后台-AGBE/…, language, strict/quality rules).

Generate test cases from doc + images + rules (MCP or script):

• MCP: Call tool generate_testcases with document_path: samples/doc-example.md, rules: content from samples/rules-example.txt, and optionally output_format: "csv" or "markdown".

• Script: DOC_PATH=samples/doc-example.md RULES_PATH=samples/rules-example.txt OUTPUT_FORMAT=both npm run generate-testcases (set OPENAI_API_KEY for LLM generation).

Project structure

mcp-manual-testcases/
├── src/
│   └── index.ts       # MCP server + generate_testcases tool
├── samples/
│   ├── doc-example.md      # Sample requirement (references Images/)
│   ├── images/             # Prototype images (image1.png …)
│   ├── rules-example.txt   # Sample rules
│   ├── generated-testcases.md
│   └── generated-testcases.csv
├── dist/              # Build output (after npm run build)
├── package.json
├── tsconfig.json
└── README.md

License

ISC

depend on doc-example.md and images folder generate tescase for the feature depend on mcp and rules