Skip to main content
Glama

MCP AI Gateway

by kinhunt

MCP AI Gateway

A unified MCP (Model Context Protocol) server that enables AI assistants to intelligently select and switch between different AI models within the same conversation to complete tasks.

šŸŽÆ Core Concept

Intelligent Model Selection

The core value of MCP AI Gateway lies in enabling AI assistants to automatically choose the most suitable model based on task requirements:

  • Code Tasks: Let AI choose Claude Opus 4 for advanced code analysis

  • Quick Q&A: Let AI choose GPT-4o for fast, reliable responses

  • Creative Writing: Let AI choose GPT-5 or Gemini 2.5 Pro for superior creativity

  • Multimodal Processing: Let AI choose vision-capable models like GPT-4o

Workflow Example

Within a single conversation, AI might:

  1. Use Claude Opus 4 to analyze complex code logic

  2. Use GPT-5 to generate creative solutions

  3. Use Gemini 2.5 Pro for quick verification

All achieved through a unified MCP tool, no manual switching required!

šŸš€ Quick Start

1. Installation and Configuration

No installation needed, use directly with npx:

npx mcp-ai-gateway

2. Claude Desktop Configuration

Add to your Claude Desktop MCP configuration:

{ "mcpServers": { "ai-gateway": { "command": "npx", "args": ["mcp-ai-gateway"], "env": { "API_FORMAT": "openai", "API_KEY": "your-api-key-here", "API_ENDPOINT": "https://api.openai.com/v1", "DEFAULT_MODEL": "gpt-4o", "DESCRIPTION": "Available models:\n- gpt-5: Latest OpenAI model with superior reasoning and creativity\n- gpt-4o: Multimodal model with fast responses\n- claude-opus-4: World's best coding model with extended thinking" } } } }

3. Start Using Immediately

After configuration, restart Claude Desktop and you can interact with AI like this:

"Please use the most suitable model to analyze this code, then use another model to generate test cases"

AI will automatically select appropriate models for different subtasks!

šŸ“– Detailed Configuration Guide

Environment Variables

Variable

Required

Description

Example

API_FORMAT

āœ…

API format

openai

or

anthropic

API_KEY

āœ…

API key

sk-...

API_ENDPOINT

⚪

Custom endpoint

https://api.openai.com/v1

DEFAULT_MODEL

⚪

Default model

gpt-4o

DESCRIPTION

⚪

Custom model description

See examples below

REQUEST_TIMEOUT

⚪

HTTP request timeout in seconds

60

(default)

API Format Support

OpenAI Format

Supports OpenAI, Azure OpenAI, and various OpenAI API-compatible services:

{ "API_FORMAT": "openai", "API_KEY": "sk-your-openai-key", "API_ENDPOINT": "https://api.openai.com/v1" }

Anthropic Format

Direct support for Anthropic Claude models:

{ "API_FORMAT": "anthropic", "API_KEY": "sk-ant-your-anthropic-key", "API_ENDPOINT": "https://api.anthropic.com", "ANTHROPIC_VERSION": "2023-06-01" }

OpenRouter Format

Access 400+ AI models through OpenRouter's unified API:

{ "API_FORMAT": "openai", "API_KEY": "sk-or-your-openrouter-key", "API_ENDPOINT": "https://openrouter.ai/api/v1", "DEFAULT_MODEL": "anthropic/claude-3.5-sonnet", "DESCRIPTION": "OpenRouter models:\n- anthropic/claude-3.5-sonnet: Latest Claude with enhanced reasoning\n- openai/gpt-4o: GPT-4o with multimodal capabilities\n- google/gemini-pro-1.5: Google's advanced Gemini model\n- meta-llama/llama-3.1-405b: Meta's largest Llama model" }

Custom Description Examples

Through the DESCRIPTION environment variable, you can provide detailed model selection guidance for AI:

export DESCRIPTION="Available AI models and their strengths: 🧠 Reasoning & Analysis: - claude-opus-4: World's best coding model with 72.5% on SWE-Bench - gpt-5: Latest model with deep reasoning capabilities and lowest error rates ⚔ Speed & Efficiency: - gpt-4o: Fast multimodal responses with near-instant processing - claude-sonnet-4: Quick processing with extended thinking capabilities - gemini-2.5-flash: Ultra-fast responses for simple queries šŸŽØ Creativity & Writing: - gpt-5: Superior creative writing and content generation - gemini-2.5-pro: Excellent balance of creativity and factual accuracy - claude-opus-4: Advanced reasoning for complex creative tasks šŸ’” Choose the model that best fits your specific task requirements!"

šŸ› ļø Advanced Configuration

Enterprise Proxy Support

export HTTP_PROXY=http://your-proxy:8080 export HTTPS_PROXY=https://your-proxy:8080

Default Parameter Settings

export DEFAULT_TEMPERATURE=0.7 export DEFAULT_MAX_TOKENS=2000 export OPENAI_ORGANIZATION=org-your-org-id # OpenAI only export REQUEST_TIMEOUT=60 # HTTP timeout in seconds (default: 60)

Multi-Provider Configuration Example

You can configure multiple MCP AI Gateway instances to connect to different providers:

{ "mcpServers": { "openai-gateway": { "command": "npx", "args": ["mcp-ai-gateway"], "env": { "API_FORMAT": "openai", "API_KEY": "sk-your-openai-key", "DESCRIPTION": "OpenAI models: GPT-5, GPT-4o, GPT-4.5" } }, "claude-gateway": { "command": "npx", "args": ["mcp-ai-gateway"], "env": { "API_FORMAT": "anthropic", "API_KEY": "sk-ant-your-key", "DESCRIPTION": "Anthropic models: Claude Opus 4, Claude Sonnet 4" } }, "openrouter-gateway": { "command": "npx", "args": ["mcp-ai-gateway"], "env": { "API_FORMAT": "openai", "API_KEY": "sk-or-your-openrouter-key", "API_ENDPOINT": "https://openrouter.ai/api/v1", "DEFAULT_MODEL": "anthropic/claude-3.5-sonnet", "DESCRIPTION": "400+ models via OpenRouter:\n- anthropic/claude-3.5-sonnet: Enhanced reasoning\n- openai/gpt-4o: Multimodal capabilities\n- google/gemini-pro-1.5: Advanced Gemini\n- meta-llama/llama-3.1-405b: Largest open model\n- Cost-effective with automatic fallbacks" } } } }

šŸ”§ Technical Features

  • šŸ”Œ Plug & Play: Use directly via npx, no installation required

  • 🌐 Multi-API Support: OpenAI, Anthropic, OpenRouter (400+ models), custom endpoints

  • šŸ—ļø Extensible Architecture: Easy to add new API format support

  • šŸ›”ļø Enterprise Ready: Proxy support, error handling, secure authentication

  • ⚔ High Performance: Direct HTTP calls, no additional overhead

  • šŸ“ Fully Typed: Written in TypeScript, type-safe

  • šŸ’° Cost Optimization: OpenRouter integration with automatic fallbacks

šŸŽÆ Use Cases

1. Development Workflow

  • Use Claude Opus 4 for code review and optimization suggestions

  • Use GPT-5 for technical documentation generation

  • Use fast models for syntax checking

2. Content Creation

  • Use creative models for draft generation

  • Use analytical models for content optimization

  • Use fast models for proofreading

3. Research & Analysis

  • Use reasoning models for complex data analysis

  • Use specialized models for report generation

  • Use fast models for summary generation

4. Third-Party Model Access in AI Clients

Access premium models through official AI clients:

  • Claude Desktop with OpenAI Models: Use your OpenAI API key to access GPT-5, GPT-4o in Claude Desktop interface

  • OpenRouter Integration: Access 400+ models through one API with automatic fallbacks and cost optimization

  • Third-Party API Integration: Connect expensive or specialized models (like Claude Opus 4) through custom endpoints

  • Cost Optimization: Use cheaper third-party API providers while maintaining the familiar Claude Desktop/Gemini CLI experience

  • Model Comparison: Test different providers' implementations of the same model within one interface

  • Enterprise Solutions: Access internal or fine-tuned models through your organization's API gateway

Example Configuration for accessing OpenAI models in Claude Desktop:

{ "mcpServers": { "openai-access": { "command": "npx", "args": ["mcp-ai-gateway"], "env": { "API_FORMAT": "openai", "API_KEY": "sk-your-openai-key", "API_ENDPOINT": "https://api.openai.com/v1", "DESCRIPTION": "Access OpenAI's latest models:\n- gpt-5: Most advanced reasoning\n- gpt-4o: Multimodal capabilities\n- Compare with Claude's built-in models" } } } }

Example Configuration for OpenRouter access in Claude Desktop:

{ "mcpServers": { "openrouter-access": { "command": "npx", "args": ["mcp-ai-gateway"], "env": { "API_FORMAT": "openai", "API_KEY": "sk-or-your-openrouter-key", "API_ENDPOINT": "https://openrouter.ai/api/v1", "DEFAULT_MODEL": "anthropic/claude-3.5-sonnet", "DESCRIPTION": "Access 400+ models via OpenRouter:\n- Choose from OpenAI, Anthropic, Google, Meta models\n- Automatic cost optimization and fallbacks\n- Unified pricing and billing across providers\n- Real-time model availability and performance" } } } }

šŸ“š API Reference

chat_completion Tool Parameters

Parameter

Type

Description

model

string

Specify the model to use

messages

array

Array of conversation message objects

temperature

number

Control randomness (0-2)

max_tokens

number

Maximum output length

stream

boolean

Whether to stream output

top_p

number

Nucleus sampling parameter

frequency_penalty

number

Frequency penalty

presence_penalty

number

Presence penalty

stop

string/array

Stop sequences

response_format

object

NEW

: Format of the response (OpenAI only)

Response Format Support

The response_format parameter enables structured outputs from OpenAI-compatible models:

// JSON object mode { "response_format": { "type": "json_object" } } // JSON schema mode (with strict validation) { "response_format": { "type": "json_schema", "json_schema": { "name": "user_profile", "strict": true, "schema": { "type": "object", "properties": { "name": {"type": "string"}, "age": {"type": "number"}, "email": {"type": "string", "format": "email"} }, "required": ["name", "age"] } } } }

Note: This parameter is only supported for OpenAI-format APIs and will be ignored for Anthropic-format requests.

šŸ”§ Troubleshooting

Common Issues

"Cannot find module" Errors with npx

This error typically occurs when using npx mcp-ai-gateway due to incomplete dependency downloads in the npx cache. You might see errors like:

  • Cannot find module 'node_modules/es-set-tostringtag/index.js'

  • Cannot find module 'axios'

  • Cannot find module '@modelcontextprotocol/sdk'

Solution:

# Clear npx cache completely rm -rf ~/.npm/_npx npm cache clean --force # Then retry npx mcp-ai-gateway

Why this happens: npx sometimes fails to download all transitive dependencies, especially on slower network connections or when the package has many dependencies. Clearing the cache forces a fresh, complete download.

API Key Not Found

Make sure you've set the required environment variables in your MCP client configuration:

{ "env": { "API_KEY": "your-api-key-here", "API_FORMAT": "openai" } }

Connection Timeouts

If you're experiencing timeout issues, increase the request timeout:

{ "env": { "REQUEST_TIMEOUT": "120" } }

šŸ“š Documentation

Comprehensive guides and tutorials for getting the most out of MCP AI Gateway:

Getting Started

Advanced Workflows

Use Cases & Examples

  • Client integration tutorials

  • Enterprise deployment guides

  • Cost optimization strategies

  • Workflow automation patterns

More tutorials and guides coming soon! Check back regularly or

šŸ“– Documentation & Guides

Comprehensive how-to guides for getting the most out of MCP AI Gateway:

Integration Guides

Advanced Workflows

Coming Soon

  • How to Build AI-Powered Code Review Pipelines

  • How to Set Up Cost-Effective Multi-Team AI Access

  • How to Create Custom Model Selection Strategies

  • Enterprise Deployment and Security Best Practices

šŸ¤ Contributing

Issues and Pull Requests are welcome!

šŸ“„ License

MIT License - See LICENSE file for details


Make AI assistants smarter at choosing models and boost your productivity! šŸš€

Related MCP Servers

  • -
    security
    F
    license
    -
    quality
    Facilitates interaction and context sharing between AI models using the standardized Model Context Protocol (MCP) with features like interoperability, scalability, security, and flexibility across diverse AI systems.
    Last updated -
    1
  • A
    security
    A
    license
    A
    quality
    An MCP server that enables AI applications to access 20+ model providers (including OpenAI, Anthropic, Google) through a unified interface for text and image generation.
    Last updated -
    2
    23
    MIT License
  • -
    security
    F
    license
    -
    quality
    A service that converts OpenAPI specifications into MCP tools, enabling AI assistants to interact with your API endpoints through natural language.
    Last updated -
  • A
    security
    A
    license
    A
    quality
    An MCP server that retrieves and provides AI model information from OpenAI, Anthropic, and Google through a unified interface.
    Last updated -
    3
    32
    MIT License

View all related MCP servers

MCP directory API

We provide all the information about MCP servers via our MCP API.

curl -X GET 'https://glama.ai/api/mcp/v1/servers/kinhunt/mcp-ai-gateway'

If you have feedback or need assistance with the MCP directory API, please join our Discord server