Claude Code Prompt Engineer
# Claude Code Prompt Engineer MCP Server
An advanced Model Context Protocol (MCP) server that intelligently engineers and optimizes prompts for Claude Code, with interactive refinement capabilities and automatic optimization.
## Features
### š **Intelligent Prompt Engineering**
- Automatically detects programming language and task type
- Optimizes prompts specifically for Claude Code's capabilities
- Leverages knowledge of Claude Code's tools and workflows
### š¤ **Interactive Refinement**
- Q&A-based prompt clarification system
- Context-aware question generation
- Session-based refinement process
### ā” **Automatic Optimization**
- Language-specific prompt enhancement
- Task complexity detection
- Claude Code capability awareness
- **No external API dependencies** - works entirely with built-in capabilities
## Installation & Setup
### 1. Install Dependencies
```bash
npm install
```
### 2. Configure Claude Code
Create or update your `.mcp.json` configuration file:
```json
{
"mcpServers": {
"prompt-engineer": {
"command": "node",
"args": ["/absolute/path/to/cc_peng_mcp/index.ts"],
"env": {}
}
}
}
```
### 3. Restart Claude Code
```bash
# Exit current session and start new one
claude
```
### 4. Verify Installation
```bash
/mcp
```
You should see `prompt-engineer` listed as connected.
## Usage with Claude Code
### šÆ **Auto-Optimization (Primary Usage)**
Simply use the tools directly - no setup required:
```
Use the auto_optimize tool to analyze: "fix my React app performance issues"
```
### **Interactive Mode for Complex Requests**
```
Use the engineer_prompt tool with interactive=true for: "create a new authentication system"
```
### **Direct Prompt Engineering**
```
Use the engineer_prompt tool with prompt="optimize my database queries" and language="python"
```
## Available Tools
### `auto_optimize` ā **Primary Tool**
Automatically detects and optimizes natural language text:
- `text` (required): Your natural language text/request
- `context` (optional): Additional project context
- `interactive` (optional): Force interactive questioning
### `engineer_prompt`
Manual prompt engineering with specific parameters:
- `prompt` (required): The raw user prompt that needs engineering
- `language` (optional): Programming language (auto-detected if not provided)
- `context` (optional): Additional project context
- `interactive` (optional): Enable Q&A refinement process
### `answer_questions`
Used to provide answers during interactive sessions:
- `sessionId` (required): Session ID from interactive mode
- `answers` (required): Array of answers to the questions
## How It Works
### 1. **Natural Language Detection**
The system automatically identifies when your text is natural language that needs optimization by detecting:
- **Request patterns**: "help me", "can you", "i need", "i want"
- **Problem descriptions**: "issue", "bug", "not working", "broken"
- **Task indicators**: "create", "build", "fix", "optimize"
- **Questions**: "how do", "what is", "why is"
- **Conversational markers**: "i'm", "i think", "not sure"
### 2. **Built-in Intelligence**
- **No API calls required** - uses pattern matching and rule-based optimization
- **Language detection** via regex patterns for 10+ programming languages
- **Task type detection** (debug, test, refactor, explain, architecture)
- **Complexity analysis** based on text length and technical indicators
### 3. **Claude Code Optimization**
Prompts are optimized knowing Claude Code's capabilities:
- File system operations (Read, Write, Edit)
- Code search (Grep, Glob patterns)
- Git operations and workflow management
- Todo list management for complex tasks
- Browser automation via Playwright MCP
- GitHub integration
### 4. **Interactive Refinement**
When needed, the system:
- Generates clarifying questions based on your prompt
- Creates a session to track the conversation
- Refines the prompt based on your answers
- Delivers a highly optimized final prompt
## Example Workflows
### šÆ **Auto-Optimization Example**
**Input:**
```
"my website is slow and users are complaining"
```
**Auto-Optimized Output:**
```
**Task:** Debug and fix the following issue:
my website is slow and users are complaining
**Requirements:**
- Use file search tools (Grep/Glob) to locate relevant code
- Read and analyze the problematic files
- Identify the root cause and implement a fix
- Test the solution if possible
**Ready to proceed with this task?**
```
### š¤ **Interactive Mode Example**
**Input:** "make my app better"
**Questions Generated:**
1. What specific aspect needs improvement (performance, readability, maintainability)?
2. What technology stack or programming language are you using?
3. Are there any constraints or requirements I should know about?
**After Answers:** Creates detailed, structured prompt with your specific requirements.
## Benefits
ā
**No Setup Required**: Works immediately after MCP configuration
ā
**No API Keys**: Uses only built-in Claude Code capabilities
ā
**Context Aware**: Understands your project and requirements
ā
**Interactive**: Asks clarifying questions when needed
ā
**Systematic**: Structures complex requests properly
ā
**Fast**: No external API calls for instant optimization
## Requirements
- Node.js 16+
- TypeScript support (tsx recommended)
- Claude Code CLI
- **No external API keys required**
## Technical Details
### Supported Languages
- JavaScript/Node.js, TypeScript, Python, Java, C++, Rust, Go, PHP, Ruby, C#/.NET
### Task Types
- **Debug**: Bug fixing and error resolution
- **Test**: Unit testing and test creation
- **Refactor**: Code improvement and restructuring
- **Explain**: Code explanation and documentation
- **Architecture**: System design and patterns
- **Code**: General coding tasks (default)
### Architecture
- **Pure pattern matching** - no machine learning models
- **Rule-based optimization** - deterministic and fast
- **Session management** - for interactive conversations
- **Built-in language detection** - regex-based patterns
## License
MIT License - feel free to modify and distribute.
## Installation Guide
See [INSTALL.md](INSTALL.md) for detailed installation instructions, troubleshooting, and advanced configuration options.TDQS
Scored across 4 tools
There is some overlap between tools, particularly 'auto_optimize' and 'engineer_prompt' both involving prompt optimization, which could cause confusion. However, 'answer_questions' and 'ask_clarification' are clearly distinct, and descriptions help differentiate the optimization tools by specifying different use cases (natural language text vs. general prompt engineering).
The tools follow a consistent verb_noun pattern throughout (e.g., answer_questions, ask_clarification, auto_optimize, engineer_prompt), with all using snake_case. There is a minor deviation in that 'auto_optimize' and 'engineer_prompt' are more descriptive phrases rather than simple verb_noun combos, but the overall naming is predictable and readable.
With 4 tools, this is a well-scoped set for a Claude Code prompt engineering server. Each tool appears to serve a distinct purpose in the workflow (clarifying, optimizing, engineering), and the count is neither too thin nor too heavy, fitting the domain appropriately.
The tool surface covers key aspects of prompt engineering: gathering requirements (ask_clarification), providing feedback (answer_questions), and optimization (auto_optimize and engineer_prompt). A minor gap might be the lack of a tool for testing or evaluating prompts, but the core workflow is well-covered, and agents can likely work around this.