Humanitarian Negotiation MCP Server

# Deploying as a Universal MCP Server ## 📌 Current Status Tu MCP **ya es agnóstico a LLMs** porque usa `FastMCP`, que es parte del ecosistema MCP estándar de Anthropic. Sin embargo, aquí hay varias formas de hacerlo verdaderamente universal y accesible desde cualquier aplicación. --- ## 🎯 Opciones de Deployment ### Opción 1: MCP Server Estándar (Recomendado - Ya lo tienes) Tu servidor actual ya es universal. Puede usarse con: - ✅ **Claude Desktop** (ya configurado) - ✅ **Claude Web** (próximamente) - ✅ **Cline** (VSCode extension) - ✅ **Continue** (IDE extension) - ✅ **Cursor IDE** - ✅ Cualquier cliente MCP compatible **Configuración en otros clientes:** ```json // Ejemplo para Cursor IDE, Continue, o Cline { "mcpServers": { "humanitarian-negotiation": { "command": "python", "args": ["/ruta/absoluta/a/humanitarian_negotiation_mcp.py"] } } } ``` --- ### Opción 2: Servidor HTTP (Más Universal) Expone el MCP como API REST para que cualquier aplicación lo use. #### Pasos: **1. Crear un wrapper HTTP:** ```python # http_mcp_server.py from fastapi import FastAPI, HTTPException from fastapi.middleware.cors import CORSMiddleware import uvicorn import asyncio import json from humanitarian_negotiation_mcp import ( create_island_of_agreement, analyze_icebergs, analyze_stakeholders, leverage_stakeholder_influence, get_negotiation_guide ) app = FastAPI( title="Humanitarian Negotiation MCP API", description="Universal REST API for humanitarian negotiation analysis", version="1.0.0" ) # Enable CORS for universal access app.add_middleware( CORSMiddleware, allow_origins=["*"], allow_credentials=True, allow_methods=["*"], allow_headers=["*"], ) # Health check @app.get("/health") async def health(): return {"status": "ok", "service": "humanitarian-negotiation-mcp"} # Tool endpoints @app.post("/tools/island-of-agreement") async def api_island_of_agreement(request: dict): try: result = create_island_of_agreement( situation_description=request.get("situation_description"), organization_name=request.get("organization_name"), counterpart_name=request.get("counterpart_name"), additional_context=request.get("additional_context"), response_format=request.get("response_format", "markdown"), detail_level=request.get("detail_level", "detailed") ) return {"success": True, "data": result} except Exception as e: raise HTTPException(status_code=400, detail=str(e)) @app.post("/tools/analyze-icebergs") async def api_analyze_icebergs(request: dict): try: result = analyze_icebergs( organization_positions=request.get("organization_positions", []), organization_reasoning=request.get("organization_reasoning", []), organization_motives=request.get("organization_motives", []), counterpart_positions=request.get("counterpart_positions", []), counterpart_reasoning=request.get("counterpart_reasoning", []), counterpart_motives=request.get("counterpart_motives", []), response_format=request.get("response_format", "markdown"), detail_level=request.get("detail_level", "detailed") ) return {"success": True, "data": result} except Exception as e: raise HTTPException(status_code=400, detail=str(e)) @app.post("/tools/analyze-stakeholders") async def api_analyze_stakeholders(request: dict): try: result = analyze_stakeholders( context=request.get("context"), stakeholders=request.get("stakeholders", []), response_format=request.get("response_format", "markdown"), detail_level=request.get("detail_level", "detailed") ) return {"success": True, "data": result} except Exception as e: raise HTTPException(status_code=400, detail=str(e)) @app.post("/tools/leverage-influence") async def api_leverage_influence(request: dict): try: result = leverage_stakeholder_influence( target_stakeholder_name=request.get("target_stakeholder_name"), stakeholders_analysis_json=request.get("stakeholders_analysis_json"), response_format=request.get("response_format", "markdown") ) return {"success": True, "data": result} except Exception as e: raise HTTPException(status_code=400, detail=str(e)) @app.get("/tools/guide") async def api_guide(): try: result = get_negotiation_guide() return {"success": True, "data": result} except Exception as e: raise HTTPException(status_code=400, detail=str(e)) # Tool discovery endpoint @app.get("/tools") async def list_tools(): return { "tools": [ { "name": "humanitarian_create_island_of_agreement", "description": "Creates IoA table with contested/agreed facts and convergent/divergent norms", "endpoint": "/tools/island-of-agreement" }, { "name": "humanitarian_analyze_icebergs", "description": "Compares parties' positions, reasoning, and motives", "endpoint": "/tools/analyze-icebergs" }, { "name": "humanitarian_analyze_stakeholders", "description": "Characterizes and prioritizes stakeholders", "endpoint": "/tools/analyze-stakeholders" }, { "name": "humanitarian_leverage_stakeholder_influence", "description": "Develops tactics to influence target stakeholders", "endpoint": "/tools/leverage-influence" }, { "name": "humanitarian_negotiation_guide", "description": "Comprehensive guide to all methodologies", "endpoint": "/tools/guide" } ] } if __name__ == "__main__": uvicorn.run( app, host="0.0.0.0", port=8000, log_level="info" ) ``` **2. Instalar dependencias adicionales:** ```bash pip install fastapi uvicorn ``` **3. Ejecutar el servidor:** ```bash python http_mcp_server.py ``` **4. El servidor estará disponible en:** - Local: `http://localhost:8000` - Documentación interactiva: `http://localhost:8000/docs` --- ### Opción 3: Docker Container (Máxima Portabilidad) Para que cualquiera pueda ejecutar tu MCP en cualquier sistema. **1. Crear Dockerfile:** ```dockerfile FROM python:3.10-slim WORKDIR /app # Copy requirements COPY requirements_mcp.txt . # Install dependencies RUN pip install --no-cache-dir -r requirements_mcp.txt # Copy MCP server COPY humanitarian_negotiation_mcp.py . # Expose port for HTTP server EXPOSE 8000 # Run the MCP server CMD ["python", "humanitarian_negotiation_mcp.py"] # Alternative: Run HTTP wrapper # CMD ["python", "http_mcp_server.py"] ``` **2. Crear .dockerignore:** ``` __pycache__ *.pyc *.pyo .pytest_cache .git .gitignore *.md .env ``` **3. Construir la imagen:** ```bash docker build -t humanitarian-negotiation-mcp:1.0 . ``` **4. Ejecutar el contenedor:** ```bash # Opción 1: MCP directo (con stdout) docker run -it humanitarian-negotiation-mcp:1.0 # Opción 2: HTTP server docker run -p 8000:8000 humanitarian-negotiation-mcp:1.0 ``` --- ### Opción 4: Publicar en Package Repository Para máxima distribución: PyPI, conda, etc. **1. Estructura del proyecto:** ``` humanitarian-negotiation-mcp/ ├── src/ │ └── humanitarian_negotiation_mcp/ │ ├── __init__.py │ ├── server.py │ └── http_wrapper.py ├── setup.py ├── pyproject.toml ├── requirements.txt └── README.md ``` **2. setup.py:** ```python from setuptools import setup, find_packages setup( name="humanitarian-negotiation-mcp", version="1.0.0", description="MCP server for humanitarian negotiation analysis", author="Your Name", author_email="your.email@example.com", url="https://github.com/yourusername/humanitarian-negotiation-mcp", packages=find_packages(where="src"), package_dir={"": "src"}, python_requires=">=3.10", install_requires=[ "anthropic>=0.18.0", "mcp>=0.9.0", "pydantic>=2.0.0", "httpx>=0.24.0", "python-dateutil>=2.8.0", ], entry_points={ "console_scripts": [ "humanitarian-mcp=humanitarian_negotiation_mcp.server:main", "humanitarian-mcp-http=humanitarian_negotiation_mcp.http_wrapper:main", ], }, classifiers=[ "Programming Language :: Python :: 3", "Programming Language :: Python :: 3.10", "Programming Language :: Python :: 3.11", "Programming Language :: Python :: 3.12", "Programming Language :: Python :: 3.13", "License :: OSI Approved :: MIT License", "Intended Audience :: Developers", "Topic :: Software Development :: Libraries", ], ) ``` **3. Publicar en PyPI:** ```bash pip install build twine python -m build twine upload dist/* ``` **Uso después de publicar:** ```bash pip install humanitarian-negotiation-mcp humanitarian-mcp # Ejecuta el servidor directamente ``` --- ### Opción 5: Integración con LLMs Específicos Para usar directamente con APIs de LLMs (OpenAI, Anthropic, etc.) **1. Cliente OpenAI con MCP:** ```python from openai import OpenAI import json client = OpenAI(api_key="your-api-key") # Define las herramientas MCP para OpenAI tools = [ { "type": "function", "function": { "name": "humanitarian_create_island_of_agreement", "description": "Creates IoA analysis with contested/agreed facts and convergent/divergent norms", "parameters": { "type": "object", "properties": { "situation_description": {"type": "string"}, "organization_name": {"type": "string"}, "counterpart_name": {"type": "string"}, "additional_context": {"type": "string"}, "response_format": {"type": "string", "enum": ["markdown", "json"]}, "detail_level": {"type": "string", "enum": ["concise", "detailed"]} }, "required": ["situation_description", "organization_name", "counterpart_name"] } } } ] messages = [ { "role": "user", "content": "Analyze this negotiation: UN agency negotiating with regional government..." } ] response = client.chat.completions.create( model="gpt-4", messages=messages, tools=tools, tool_choice="auto" ) ``` --- ## 🚀 Recomendación: Camino a Universal Para maximizar alcance y compatibilidad, recomiendo esta secuencia: ### 1️⃣ Corto Plazo (Ya hecho) - ✅ MCP estándar funcional - ✅ Compatible con Claude Desktop - ✅ Documentación clara ### 2️⃣ Mediano Plazo - 📝 Crear versión HTTP con FastAPI - 📝 Documentación Swagger/OpenAPI - 📝 Ejemplos para múltiples clientes ### 3️⃣ Largo Plazo - 📦 Publicar en PyPI - 🐳 Docker container - 🌐 Hosting en nube (AWS Lambda, Google Cloud Run, etc.) --- ## 🔗 Cómo Usar en Diferentes Plataformas ### Claude Desktop ✅ (Ya configurado) ```json { "mcpServers": { "humanitarian-negotiation": { "command": "python", "args": ["/ruta/a/humanitarian_negotiation_mcp.py"] } } } ``` ### Cline (VSCode) Añade la misma configuración a `.cline/mcp-config.json` ### Continue IDE ```json { "contextProviders": [ { "name": "mcp", "params": { "mcpServers": { "humanitarian-negotiation": { "command": "python", "args": ["/ruta/a/humanitarian_negotiation_mcp.py"] } } } } ] } ``` ### Cursor IDE Stalkear la configuración de Claude Desktop (usa el mismo archivo) ### API REST (Tu propia app) ```bash curl -X POST http://localhost:8000/tools/island-of-agreement \ -H "Content-Type: application/json" \ -d '{ "situation_description": "...", "organization_name": "...", "counterpart_name": "..." }' ``` --- ## 📊 Comparativa de Opciones | Opción | Facilidad | Universalidad | Escalabilidad | Complejidad | |--------|-----------|--------------|---------------|------------| | MCP Estándar | ⭐⭐⭐⭐⭐ | ⭐⭐⭐⭐ | ⭐⭐⭐ | Baja | | HTTP API | ⭐⭐⭐⭐ | ⭐⭐⭐⭐⭐ | ⭐⭐⭐⭐ | Media | | Docker | ⭐⭐⭐ | ⭐⭐⭐⭐⭐ | ⭐⭐⭐⭐ | Media | | PyPI | ⭐⭐⭐ | ⭐⭐⭐⭐⭐ | ⭐⭐⭐⭐⭐ | Alta | | LLM Integration | ⭐⭐⭐⭐ | ⭐⭐⭐⭐⭐ | ⭐⭐⭐⭐ | Alta | --- ## 💡 Conclusión Tu MCP **ya es agnóstico** y puede usarse con cualquier cliente compatible. Para hacerlo "verdaderamente universal": 1. **Inmediato**: Usa con Cline, Continue, Cursor 2. **Próxima semana**: Crea wrapper HTTP 3. **Este mes**: Publica en PyPI y crea Docker image 4. **Largo plazo**: Hospédalo en nube como servicio ¿Quieres que implemente alguna de estas opciones?