Skip to main content
Glama
parigihelori352-cyber

model-gateway

by parigihelori352-cyber
OverviewSchemaRelated ServersScoreDiscussions
Python
Remote

model-gateway — Define MCP Tools in JSON, Not Python

One MCP server, any model, any role. Add a new LLM-powered tool in 5 minutes by writing a JSON block — zero Python code.

Why

Every MCP server follows the same pattern: define a tool schema, write a handler, call an LLM, return the result. The only thing that changes between tools is the system prompt, model, and input schema. So why write Python for each one?

model-gateway separates the what from the how:

config.json  ←──  WHAT: providers, models, tools, system prompts (you edit this)
    │
    ▼
server.py    ←──  HOW: MCP protocol, API routing, multimodal encoding (never touch this)
    │
    ▼
Claude Code  ←──  sees N tools, each powered by a different model

Related MCP server: Dynamic MCP Server

Quick Start

# 1. Install
pip install -e .

# 2. Set API keys (or put them in ~/.claude/settings.json → env)
export OPENROUTER_API_KEY="sk-or-v1-xxx"

# 3. Define a tool — add this to config.json → capabilities:
# {
#   "tool": "my_summarizer",
#   "description": "Summarize text concisely",
#   "provider": "openrouter",
#   "model": "openai/gpt-4o-mini",
#   "system_prompt": "Summarize the text. Output only the summary.",
#   "input_schema": {
#     "type": "object",
#     "properties": {
#       "text": { "type": "string", "description": "Text to summarize" }
#     },
#     "required": ["text"]
#   }
# }

# 4. Restart Claude Code — my_summarizer appears automatically.

How It Works

Providers (API backends)

Configure once in config.json → providers. Each provider knows its base_url and which env var holds the key:

Provider

Env Var

Used for

OpenRouter

OPENROUTER_API_KEY

GPT-5.4, Gemini, Claude… (any model on OpenRouter)

OpenAI

OPENAI_API_KEY

Direct GPT-4o access

Ark (Volcano)

ARK_API_KEY

Doubao vision models

Add a new provider in one JSON block — no code needed.

Tools (capabilities)

Every tool is a JSON object with:

Field

Purpose

tool

MCP tool name (snake_case)

description

Shown to Claude Code

provider

Which provider to route through

model

Model ID (OpenRouter format, e.g. openai/gpt-5.4)

system_prompt

The role prompt — this IS the tool's behavior

input_schema

JSON Schema — defines what arguments the tool accepts

accepts_images

(optional) Enable multimodal image analysis

reasoning

(optional) Enable reasoning tokens + effort control

Example: Build a code reviewer in 1 minute

{
  "tool": "code_reviewer",
  "description": "Review code for bugs and improvements",
  "provider": "openrouter",
  "model": "openai/gpt-5.4",
  "system_prompt": "You are a senior code reviewer. Find bugs, edge cases, and suggest improvements. Be specific — reference line numbers.",
  "reasoning": { "enabled": true, "default_effort": "medium" },
  "input_schema": {
    "type": "object",
    "properties": {
      "code": { "type": "string", "description": "Code to review" },
      "language": { "type": "string", "description": "Programming language" }
    },
    "required": ["code"]
  }
}

Restart Claude Code. Done. code_reviewer is now an MCP tool powered by GPT-5.4 with reasoning.

Built-in Tools

These come pre-configured as examples — delete or modify them freely:

Tool

Default Model

Purpose

vision_ask

Doubao Seed 2.1 Pro

Multimodal image analysis

gpt_plan

GPT-5.4

Architecture & implementation planning

gpt_review

GPT-5.4

Deep code review (security/correctness/architecture)

gpt_decide

GPT-5.4

Multi-option trade-off analysis

gpt_design_workflow

GPT-5.4

Claude Code Workflow script generation

gpt_translate

Gemini 2.5 Flash

High-quality translation

CLI

All tools are also available from the command line:

# Vision analysis
mg vision photo.jpg "Describe this image in detail"

# Architecture planning (GPT-5.4 + reasoning)
mg plan "Design a microservice saga pattern" --budget high

# Deep code review
mg review auth.py --focus security

# Trade-off analysis
mg decide "Redis Cluster" "Sentinel" "Cloud managed" --criteria "perf,cost,ops"

# Workflow script generation
mg workflow "Migrate auth to JWT with zero downtime" --quality exhaustive

# Translation
mg translate "Hello world" Chinese

Claude Code Integration

Add to ~/.claude/settings.json:

{
  "mcpServers": {
    "model-gateway": {
      "command": "python",
      "args": ["-m", "model_gateway.mcp_server"],
      "cwd": "/path/to/model-gateway"
    }
  }
}

Security

  • No API keys in config. All keys come from environment variables. config.json only stores api_key_env references.

  • .gitignore excludes config.json — your tool definitions stay local.

  • save() strips keys. The config manager writes only api_key_env references, never resolved keys.

Design Philosophy

  • Config is the API. The shape of a tool IS its definition. No indirection.

  • No code for new tools. If you can write a system prompt, you can build an MCP tool.

  • Zero overhead. The framework does nothing but route and serialize. Your prompts do the work.

  • Vendor freedom. Swap between OpenRouter, OpenAI, Ark, or your own provider — tools stay identical.

License

MIT

This server cannot be installed

A
license - permissive license
-
quality - not tested
C
maintenance

How are these scores calculated?

Maintenance

–Maintainers
–Response time
–Release cycle
–Releases (12mo)
Commit activity

Resources

Unclaimed servers have limited discoverability.

Looking for Admin?

If you are the server author, to access and configure the admin panel.

Latest Blog Posts

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/parigihelori352-cyber/model-gateway'

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