prompteneer by doozMen - MCP Server

Prompteneer - Agent Orchestration MCP Server

Privacy-focused agent orchestration server using Apple's Foundation Models framework for intelligent agent discovery, request analysis, and routing.

Overview

Prompteneer provides secure, on-device agent orchestration capabilities through the Model Context Protocol. Using Apple's Foundation Models with keyword-based fallback, it analyzes user requests and intelligently matches them with specialized Claude Code agents. All processing happens locally - no data is sent to external servers.

Features

Privacy-First: 100% on-device processing with Apple Intelligence
Agent Orchestration: Discover, match, and route to 40+ specialized agents
Intelligent Prompt Enhancement: Automatically rebuild prompts with contextual agent injection
Foundation Models: Native Apple Intelligence integration (macOS 26+)
Hybrid Fallback: Keyword-based analysis for macOS 15+ compatibility
Request Analysis: Intent detection, technology identification, complexity scoring
Text Analysis: Query, summarize, sentiment analysis (legacy tools)
PII Detection: Identify personally identifiable information before sharing
MCP Integration: Seamless integration with Claude Code

Installation

Prerequisites

macOS 15.0+
Swift 6.0+
Xcode Command Line Tools

Quick Install

cd prompteneer
./install.sh

The install script will:

Build in release mode
Install to ~/.swiftpm/bin
Automatically configure Claude Code (if CLI available)
Create config backup for safety

Manual Build and Install

# Build in release mode
swift build -c release

# Install to ~/.swiftpm/bin
rm -f ~/.swiftpm/bin/prompteneer
swift package experimental-install

Verify Installation

prompteneer --version
# Should output: prompteneer 0.0.1

which prompteneer
# Should output: /Users/[username]/.swiftpm/bin/prompteneer

Configuration

Option 1: Automatic (Recommended)

The install script automatically configures Claude Code. If you need to reconfigure:

claude mcp add -s user prompteneer prompteneer \
  -e PATH="$HOME/.swiftpm/bin:/usr/local/bin:/usr/bin:/bin"

Option 2: Manual Configuration

Add to ~/.claude.json:

{
  "mcpServers": {
    "prompteneer": {
      "type": "stdio",
      "command": "prompteneer",
      "args": ["--log-level", "info"],
      "env": {
        "PATH": "$HOME/.swiftpm/bin:/usr/local/bin:/usr/bin:/bin"
      }
    }
  }
}

Restart Claude Code after configuration:

claude-code restart

Available Tools

Prompteneer provides 9 powerful tools via MCP protocol:

Agent Orchestration Tools (Primary Focus):

analyze_request - Analyze user requests (intent, technologies, complexity)
discover_agents - List all available Claude Code agents
match_agents - Match agents to requests with relevance scoring
explain_routing - Explain why specific agents were recommended
optimize_prompt - Flagship: Intelligently rebuild prompts with agent injection

Legacy LLM Tools (Foundational):

query - Execute LLM queries with local models
summarize - Summarize text content
analyze_sentiment - Sentiment analysis
detect_pii - Detect personally identifiable information

1. query - Execute LLM Query

Execute a query using local AI models.

Parameters:

prompt (string, required): User query or prompt
model (string, optional): Model to use
- "apple-intelligence" (default)
- "phi-3-mini"
- "tinyllama"
max_tokens (int, optional): Maximum tokens (default: 1000)
temperature (float, optional): Sampling temperature 0.0-1.0 (default: 0.7)

Example:

{
  "name": "query",
  "arguments": {
    "prompt": "Explain quantum computing in simple terms",
    "model": "apple-intelligence",
    "max_tokens": 500
  }
}

Returns:

Response: [AI-generated response]

Model: Apple Intelligence
Tokens: 125
Inference Time: 342ms
Privacy: On-device (✓)

2. summarize - Summarize Text

Summarize long text content locally.

Parameters:

text (string, required): Content to summarize
max_length (int, optional): Maximum word count for summary

Example:

{
  "name": "summarize",
  "arguments": {
    "text": "Long article text here...",
    "max_length": 100
  }
}

Returns:

Summary: [Condensed summary]

Original Length: 1523 chars
Summary Length: 456 chars
Compression Ratio: 0.30

3. analyze_sentiment - Sentiment Analysis

Analyze emotional tone of text locally.

Parameters:

text (string, required): Content to analyze

Example:

{
  "name": "analyze_sentiment",
  "arguments": {
    "text": "I absolutely loved the new feature! It works great."
  }
}

Returns:

Sentiment: Positive 😊
Confidence: 87%

Analysis performed locally on-device.

4. detect_pii - Detect Personal Information

Scan text for personally identifiable information.

Parameters:

text (string, required): Content to scan

Example:

{
  "name": "detect_pii",
  "arguments": {
    "text": "Contact John Smith at john@example.com"
  }
}

Returns:

⚠️ PII Detected: YES

Detected Types: email, name

Warning: This text contains personally identifiable information.

5. analyze_request - Analyze User Request

Analyze user requests to understand intent, extract keywords, identify technologies, and estimate complexity using Foundation Models or keyword-based analysis.

Parameters:

request (string, required): User request or query to analyze

Example:

{
  "name": "analyze_request",
  "arguments": {
    "request": "I need to build a SwiftUI app with Core Data and Firebase"
  }
}

Returns (Foundation Models):

Request Analysis (Foundation Models):

Intent: Development
Keywords: SwiftUI, Core Data, Firebase
Technologies: Swift, Core Data, Firebase
Complexity: 0.7

Analysis performed using Apple Intelligence.

Returns (Keyword Fallback):

Request Analysis (Keyword-Based):

Intent: development
Keywords: build, swiftui, app, core, data, firebase
Technologies: swift, database, firebase
Complexity Score: 0.68
Confidence: 0.60

Analysis performed using keyword-based heuristics.

6. discover_agents - Discover Available Agents

List all available Claude Code agents from ~/.claude/agents/ directory with their capabilities, tools, and models.

Parameters: None

Example:

{
  "name": "discover_agents",
  "arguments": {}
}

Returns:

Discovered 40 agent(s):

- swift-architect
  ID: swift-architect.md
  Description: Specialized in Swift 6.0 architecture patterns, async/await, actors
  Tools: Read, Edit, Glob, Grep, Bash, MultiEdit
  Model: sonnet

- swiftui-specialist
  ID: swiftui-specialist.md
  Description: Expert in SwiftUI app development and state management
  Tools: Read, Edit, Glob, Grep, Bash, MultiEdit
  Model: sonnet

[... 38 more agents ...]

7. match_agents - Match Agents to Request

Match available agents to a user request based on relevance. Returns top N agents with relevance scores and reasoning.

Parameters:

request (string, required): User request to match against agents
top_n (int, optional): Number of top agents to return (default: 3)

Example:

{
  "name": "match_agents",
  "arguments": {
    "request": "Help me migrate my iOS app to Swift 6 concurrency with actors",
    "top_n": 3
  }
}

Returns:

Top 3 agent(s) for request:

1. swift-architect (Score: 46%)
   Description: Specialized in Swift 6.0 architecture patterns, async/await, actors
   Tools: Read, Edit, Glob, Grep, Bash, MultiEdit
   Reasoning: Keyword overlap: 46% - Matched terms in agent description and tools

2. swift-modernizer (Score: 38%)
   Description: Focused on migrating legacy code to Swift 6.0
   Tools: Read, Edit, Glob, Grep, Bash, MultiEdit, WebSearch
   Reasoning: Keyword overlap: 38% - Matched terms in agent description and tools

3. swift-developer (Score: 28%)
   Description: Expert Swift developer for implementing features
   Tools: Read, Edit, Glob, Grep, Bash, MultiEdit
   Reasoning: Keyword overlap: 28% - Matched terms in agent description and tools

Request Analysis:
- Intent: architecture
- Technologies: swift, ios, actor, concurrency
- Complexity: 0.72

8. explain_routing - Explain Agent Recommendations

Provide detailed explanations for why specific agents were recommended for a request, including matched criteria and reasoning.

Parameters:

request (string, required): User request to explain routing for
top_n (int, optional): Number of top agents to explain (default: 3)

Example:

{
  "name": "explain_routing",
  "arguments": {
    "request": "Help me write comprehensive tests for my Swift package",
    "top_n": 2
  }
}

Returns:

Routing Explanation:

Request Analysis:
- Intent: testing
- Keywords: help, write, comprehensive, tests, swift, package
- Technologies: swift, testing
- Complexity Score: 0.19
- Confidence: 0.60

Agent Recommendations:

1. swift-testing-xcode-specialist (Score: 41%)

   Why this agent?
   Keyword overlap: 41% - Matched terms in agent description and tools

   Agent capabilities:
   - Description: Expert in Swift Testing framework integration with Xcode projects
   - Tools: Read, Edit, Glob, Grep, Bash, MultiEdit
   - Model: sonnet

   Matching criteria:
   - Agent has expertise in areas matching request intent (testing)
   - Agent tools and description align with requested technologies
   - Score reflects keyword overlap and weighted matching

2. swift-testing-specialist (Score: 28%)

   [Similar detailed explanation...]

Matching Algorithm:
- Keyword-based analysis extracts intent, technologies, and keywords
- Agents scored by keyword overlap with name (3x weight), description (2x), and tools (1x)
- Top 2 agents selected by relevance score
- Foundation Models used for analysis when available (macOS 26+)

9. optimize_prompt - Intelligent Prompt Enhancement with Agent Injection

Flagship Feature: Intelligently rebuild prompts with contextual agent injection. Analyzes intent, matches relevant agents, and uses Foundation Models or template-based approach to inject @agent mentions directly into your prompt - making it immediately actionable.

💡 Key Benefits

Cost Reduction - Better structured prompts reduce retries and token usage by eliminating ambiguity and providing clear direction from the start
Context Preservation - Maintains clarity and focus throughout conversations with numbered agent workflows and explicit expertise areas
Better Agent Routing - Explicit @agent-{name} mentions guide Claude Code to invoke the right specialists immediately
Privacy-Preserving - All processing happens on-device with Foundation Models (macOS 26+) or keyword-based analysis (macOS 15+)
Intelligent Workflow - Agents organized in logical sequences (architect → developer → tester) for complex multi-step tasks

Parameters:

prompt (string, required): The prompt to optimize and enhance
context (string, optional): Additional context about the prompt's purpose
max_agents (int, optional): Maximum number of agents to inject (default: 3, max: 10)
inject_agents (bool, optional): Whether to inject agent mentions (default: true)

Basic Example:

{
  "name": "optimize_prompt",
  "arguments": {
    "prompt": "I need to refactor my legacy Swift code",
    "max_agents": 2
  }
}

Returns:

# Enhanced Prompt

I need to refactor my legacy Swift code to modern Swift 6.0 patterns.

To accomplish this:

1. Use @agent-swift-modernizer for migrating legacy code to Swift 6.0
2. Employ @agent-swift-architect for architectural guidance on concurrency patterns

## Injected Agents
- @agent-swift-modernizer
- @agent-swift-architect

## Improvements Made
- Clarified modernization goal
- Specified target Swift version
- Matched 2 relevant agents

## Optimization Method
Template-based

## Analysis
- Intent: refactoring
- Technologies: swift, code, legacy
- Complexity: 0.35

Real-World Example: Server-Side API Development

Input:

{
  "name": "optimize_prompt",
  "arguments": {
    "prompt": "Help me build a server-side Swift API with database integration",
    "inject_agents": true,
    "max_agents": 4
  }
}

Output:

# Enhanced Prompt

I need help building a server-side Swift API with database integration.

Working on production iOS app with Swift 6.0, I should:

1. Use @agent-swift-mcp-server-writer to create MCP servers using Swift 6.
2. Employ @agent-hummingbird-developer for a lightweight Swift HTTP server using Hummingbird framework for modern APIs.
3. Leverage @agent-swift-grpc-temporal-developer for Swift gRPC v2 streaming and Temporal workflows for production services.
4. Utilize @agent-swift-server for server-side Swift development leveraging tech-conf-mcp for learning guidance.

These agents will assist in developing a robust and efficient server-side API.

## Injected Agents
- @agent-swift-mcp-server-writer
- @agent-hummingbird-developer
- @agent-swift-grpc-temporal-developer
- @agent-swift-server

## Improvements Made
- Analyzed intent: testing
- Matched 4 relevant agent(s)
- Used Foundation Models for intelligent rebuild

## Optimization Method
Foundation Models

## Analysis
- Intent: testing
- Technologies: database, swift, api, server
- Complexity: 0.20

## Original Prompt
Help me build a server-side Swift API with database integration

Why This Matters

Cost Reduction: Better structured prompts reduce token usage and minimize retry cycles. Clear agent routing from the start saves API calls.

Context Preservation: Agents are injected with clear numbered workflows, maintaining clarity across long conversations.

Intelligent Matching: Foundation Models (macOS 26+) understand semantic intent, not just keywords. Falls back gracefully to template-based injection on earlier macOS versions.

Privacy-First: All analysis and agent matching happens on-device using Apple Intelligence. No external API calls.

Immediate Action: The enhanced prompt includes @agent mentions that Claude Code can act on immediately, eliminating the need for manual agent discovery.

Advanced Usage

Disable Agent Injection (optimization only):

{
  "name": "optimize_prompt",
  "arguments": {
    "prompt": "Explain quantum computing",
    "inject_agents": false
  }
}

With Context (better agent matching):

{
  "name": "optimize_prompt",
  "arguments": {
    "prompt": "Add tests to my code",
    "context": "Swift package using Swift Testing framework, not XCTest",
    "max_agents": 2
  }
}

Returns:

# Enhanced Prompt

Add comprehensive tests to my Swift package code.

Context: Swift package using Swift Testing framework, not XCTest

Recommended approach:

1. Use @agent-testing-specialist for Swift Testing framework expertise
2. Employ @agent-swift-testing-xcode-specialist for Xcode test configuration

## Injected Agents
- @agent-testing-specialist
- @agent-swift-testing-xcode-specialist

## Improvements Made
- Integrated provided context
- Specified testing framework (Swift Testing, not XCTest)
- Matched 2 testing specialists

## Optimization Method
Template-based

## Analysis
- Intent: testing
- Technologies: swift, code, testing, framework
- Complexity: 0.28

How It Works

Request Analysis: Analyzes prompt using Foundation Models or keyword-based analyzer
Agent Discovery: Scans ~/.claude/agents/ for available specialists
Smart Matching: Scores agents by relevance to intent and technologies
Intelligent Rebuild: Uses Foundation Models (preferred) or template-based approach to restructure prompt with agent workflow
Contextual Injection: Agents are integrated naturally with numbered steps explaining their expertise

Optimization Methods

Foundation Models (macOS 26+):

Semantic understanding of intent
Natural language prompt reconstruction
Contextual agent expertise descriptions
Best quality, ~100-500ms latency

Template-Based (macOS 15+):

Deterministic workflow generation
Keyword-based agent matching
Structured numbered lists
Fast and reliable, <10ms latency

Both methods provide immediate value - the enhanced prompt is ready to use with agents already injected.

Development

Build

swift build

Test

swift test

Code Formatting

# Lint
swift format lint -s -p -r Sources Tests Package.swift

# Auto-fix formatting
swift format format -p -r -i Sources Tests Package.swift

Run with Debug Logging

swift run prompteneer --log-level debug

Manual MCP Protocol Test

echo '{"jsonrpc":"2.0","method":"tools/list","id":1}' | prompteneer

Architecture

Prompteneer MCP Server
├── LocalLLMMCP.swift                  # Main entry point (@main with ArgumentParser)
├── MCPServer.swift                    # Actor-based MCP protocol handler
├── Models/
│   ├── Agent.swift                    # Agent domain models
│   ├── FoundationModelsTypes.swift    # @Generable types for structured output
│   ├── QueryOptions.swift             # Legacy LLM request parameters
│   ├── LLMResult.swift                # Legacy LLM response types
│   └── Errors.swift                   # Error definitions
└── Services/
    ├── FoundationModelsManager.swift  # Actor for Apple Intelligence
    ├── KeywordBasedAnalyzer.swift     # Actor for fallback analysis
    ├── KeywordBasedMatcher.swift      # Actor for agent matching
    ├── AgentDiscovery.swift           # Actor for agent discovery
    ├── LLMEngine.swift                # Legacy AI inference engine
    └── PIIDetector.swift              # Privacy scanner with regex

Key Architectural Patterns

Actor-Based Concurrency: All services are actors for Swift 6.0 strict concurrency
Hybrid Strategy: Foundation Models primary path with keyword-based fallback
Single-Process Design: All actors in same process, no IPC overhead
MCP Swift SDK 0.9.0+: Type-safe tool definitions with JSON schema
Graceful Degradation: Automatic fallback when Foundation Models unavailable

Current Implementation Status

Phase 1: Agent Orchestration (Complete ✅)

✅ MCP server with 9 tools (4 legacy + 5 agent orchestration)
✅ Foundation Models integration (macOS 26+)
✅ Keyword-based fallback (macOS 15+)
✅ Agent discovery from ~/.claude/agents/
✅ Request analysis with intent/technology detection
✅ Agent matching with relevance scoring
✅ Routing explanation with reasoning
✅ Intelligent prompt optimization with agent injection (Flagship Feature)
✅ Comprehensive test suite (56/56 tests passing)
✅ Actor-based Swift 6.0 concurrency
✅ Error handling and structured logging

Phase 2: Enhanced Intelligence (Future ⏳)

⏳ Foundation Models tool calling
⏳ Multi-agent workflow orchestration
⏳ Advanced capability matching
⏳ Performance optimizations
⏳ Analytics and usage tracking

See .agent-workspace/ for detailed Phase 1 documentation.

Integration with OwlPing

Prompteneer is designed to work alongside OwlPing:

OwlPing: Notifications, sessions, bidirectional communication
Prompteneer: Agent orchestration, request analysis, intelligent routing

Both run as separate MCP servers in Claude Code. Example workflow:

Use Prompteneer to analyze a request and match specialized agents
Invoke the recommended agent for task execution
Send notifications via OwlPing when tasks complete

Privacy Guarantees

✅ 100% on-device processing
✅ No network requests to external services
✅ No data collection or telemetry
✅ PII detection helper for data sanitization
✅ Open-source and auditable

Troubleshooting

Server Not Found

# Verify installation
which prompteneer
# Should output: /Users/[username]/.swiftpm/bin/prompteneer

# Verify it runs
prompteneer --version
# Should output: prompteneer 0.0.1

MCP Connection Issues

Check Claude Code logs:

tail -f ~/Library/Logs/Claude/mcp-server-prompteneer.log

Test the MCP protocol directly:

echo '{"jsonrpc":"2.0","method":"tools/list","id":1}' | prompteneer

Foundation Models Not Available

If you see "Foundation Models available: false" in logs:

Requires macOS 26.0+ for Foundation Models
Falls back automatically to keyword-based analysis on macOS 15+
Both methods provide agent orchestration capabilities

No Agents Found

# Check agents directory
ls -la ~/.claude/agents/

# Install agents using claude-agents CLI
claude-agents install swift-architect swift-developer --global

Build Errors

# Clean build
rm -rf .build
swift build

# Check Swift version
swift --version
# Requires Swift 6.0+

Contributing

This project is part of the OwlPing ecosystem. See main repository for contribution guidelines.

License

MIT License - See LICENSE file for details.

doozMen/prompteneer

Prompteneer - Agent Orchestration MCP Server

Overview

Features

Installation

Prerequisites

Quick Install

Manual Build and Install

Verify Installation

Configuration

Option 1: Automatic (Recommended)

Option 2: Manual Configuration

Available Tools

1. query - Execute LLM Query

2. summarize - Summarize Text

3. analyze_sentiment - Sentiment Analysis

4. detect_pii - Detect Personal Information

5. analyze_request - Analyze User Request

6. discover_agents - Discover Available Agents

7. match_agents - Match Agents to Request

8. explain_routing - Explain Agent Recommendations

9. optimize_prompt - Intelligent Prompt Enhancement with Agent Injection

💡 Key Benefits

Real-World Example: Server-Side API Development

Why This Matters

Advanced Usage

How It Works

Optimization Methods

Development

Build

Test

Code Formatting

Run with Debug Logging

Manual MCP Protocol Test

Architecture

Key Architectural Patterns

Current Implementation Status

Phase 1: Agent Orchestration (Complete ✅)

Phase 2: Enhanced Intelligence (Future ⏳)

Integration with OwlPing

Privacy Guarantees

Troubleshooting

Server Not Found

MCP Connection Issues

Foundation Models Not Available

No Agents Found

Build Errors

Contributing

License

Links