MRouter - LLM Gateway Router

One Endpoint, Every Model. Route, Monitor, and Failover — All From Your Terminal.

One endpoint for every AI provider — smart routing, auto failover, zero-config CLI sync. Built as a terminal-native LLM gateway.

Why MRouter?

Server Terminal Configuration - Full TUI management via SSH on remote servers. No web UI needed — configure providers, monitor stats, and switch routing directly from any terminal session
Single Configuration Point - Configure all providers once. Applications only need http://localhost:4444
Model-Based Auto-Routing - Specify a model name (e.g., gpt-4, claude-opus-4, gemini-pro) and MRouter routes to the correct provider automatically
Centralized API Key Management - All API keys stored securely in one place, no per-application configuration needed
Automatic Failover & Circuit Breaker - If one provider fails, automatically tries alternatives with built-in fault tolerance
Custom Headers Per Provider - Override headers like User-Agent per provider for version control and compatibility
CLI Tools Auto-Sync - One-click sync configuration to Claude Code, Codex, Gemini CLI, OpenCode, OpenClaw

Key Use Cases

Multi-Provider Routing for Server Applications

When deploying applications on servers that need to access multiple AI providers, MRouter acts as a unified gateway. Your applications only need to point to http://localhost:4444 — no need to configure API keys and endpoints for each provider in every application.

# All requests go to localhost:4444 — MRouter routes by model name
curl http://localhost:4444/v1/messages -d '{"model": "gpt-4", ...}'          # → OpenAI
curl http://localhost:4444/v1/messages -d '{"model": "claude-opus-4", ...}'  # → Anthropic
curl http://localhost:4444/v1/messages -d '{"model": "gemini-pro", ...}'     # → Google

Remote Server Management via SSH

MRouter's TUI works over SSH, making it ideal for managing AI routing on remote development servers or cloud instances:

ssh my-server
mrouter              # Full TUI management in your terminal

No web browser or port forwarding needed — add providers, switch routes, view stats, and monitor logs all from the terminal.

Core Features

🚀 Provider Management - Manage multiple AI providers (Anthropic, OpenAI, Google, AWS Bedrock, etc.)
🔄 Smart Routing - Model-based intelligent routing and automatic failover
🛡️ Circuit Breaker - Built-in circuit breaker for fault tolerance
📊 Token Statistics - Track token usage and API costs per provider with input/output breakdown
💰 Pricing Configuration - Configure custom pricing per provider for accurate cost tracking
🔀 Model Mappings - Map custom model names to standard provider models
📝 Request Logs - View detailed request logs with pagination and filtering
🔌 LLM Gateway Proxy - Unified proxy endpoint for all CLI tools
⚙️ Config Sync - Automatic configuration sync to CLI tools (Claude Code, Codex, Gemini CLI, OpenCode, OpenClaw)
🔧 Custom Headers - Per-provider custom header overrides (User-Agent, etc.)
🔄 Protocol Conversion - Automatic Anthropic ↔ OpenAI protocol conversion, use any client with any provider
📦 Provider Import/Export - Export and import provider configurations for easy migration
🎨 Beautiful TUI - Intuitive terminal user interface with keyboard shortcuts, works over SSH

Quick Start

Installation

# Clone the repository
git clone https://github.com/ubiomni/mrouter.git
cd mrouter

# Build and install
cargo build --release
cargo install --path .

Basic Usage

# Start TUI (recommended)
mrouter

# Or use CLI commands
mrouter list                    # List all providers
mrouter switch <provider-name>  # Switch active provider
mrouter status                  # Show current status
mrouter stats                   # Show usage statistics

# Start proxy server
mrouter proxy start
mrouter proxy stop
mrouter proxy status

TUI Keyboard Shortcuts

Global

1-5 - Switch tabs (Providers/Proxy/Logs/Stats/Settings)
q / Ctrl+C - Quit
? / F1 - Toggle help
r - Refresh

Providers Tab

↑↓ / j/k - Navigate list
Enter - Activate provider
a - Add new provider
e - Edit provider
d - Delete provider
v - View supported models
p - Configure pricing
o - Configure headers (Auth Header + Custom Headers)
m - Configure model mappings
r - Reset circuit breaker
s - Manage sync settings
I - Import providers from JSON
E - Export providers to JSON

Proxy Tab

s - Start proxy
p - Stop proxy
e - Edit proxy config

Logs Tab

↑↓ / j/k - Navigate logs
←→ / h/l - Previous/Next page
Enter / Space - View log detail
Esc / q - Back to list (from detail)
r - Refresh logs

Stats Tab

t - Toggle time range (Today/Week/Month/All)
r - Refresh statistics

Configuration

Configuration file: ~/.config/mrouter/config.toml

[proxy]
bind = "127.0.0.1"
port = 4444
takeover_mode = false

[circuit_breaker]
failure_threshold = 5
success_threshold = 2
timeout_secs = 60
half_open_timeout_secs = 30

[log]
level = "info"
file = "~/mrouter.log"

Supported Providers

Anthropic (Claude)
OpenAI (GPT-4, GPT-3.5)
Google AI Studio (Gemini)
AWS Bedrock
Azure OpenAI
Google Vertex AI
Mistral AI
Cohere
DeepSeek
xAI (Grok)
Meta (Llama)
MiniMax
Zhipu AI
Moonshot AI
Baichuan
OpenRouter
Together AI
Fireworks AI
Groq
Custom endpoints

Proxy Mode

MRouter can act as a unified LLM Gateway proxy:

Start the proxy:
```
mrouter proxy start
```
Configure your CLI tools to use http://localhost:4444 as the base URL
Switch providers in the TUI - all CLI tools will automatically use the new provider

Running as a Background Service

MRouter can run as a daemon on servers:

# Start proxy in background
nohup mrouter proxy start > /dev/null 2>&1 &

# Or use systemd (Linux)
sudo tee /etc/systemd/system/mrouter.service > /dev/null <<'EOF'
[Unit]
Description=MRouter LLM Gateway
After=network.target

[Service]
Type=simple
ExecStart=/usr/local/bin/mrouter proxy start
Restart=on-failure
User=your-username

[Install]
WantedBy=multi-user.target
EOF

sudo systemctl enable --now mrouter

Perfect for remote development servers where you SSH in and want centralized provider management. Use the TUI (mrouter) to manage providers, view logs, and monitor stats — all from your terminal.

CLI Tool Configuration Examples

Claude Code

// ~/.claude/settings.json
{
  "env": {
    "ANTHROPIC_BASE_URL": "http://localhost:4444"
  }
}

Codex

// ~/.codex/config.json
{
  "provider": "openai",
  "model": "gpt-4"
}
// ~/.codex/auth.json
{
  "openai_api_key": "any-placeholder-key"
}

Gemini CLI

// ~/.config/gemini/config.json
{
  "base_url": "http://localhost:4444"
}

OpenClaw

# ~/.openclaw/.env
BASE_URL=http://localhost:4444

Tip: Use s key in the Providers tab to auto-sync configuration to your CLI tools, so you don't have to configure them manually.

Custom Headers Per Provider

You can configure custom headers for each provider. When a custom header is set, it overrides the client's original header (e.g., User-Agent). If not set, headers are passed through from the client as-is.

Configure custom headers when adding/editing a provider in JSON format:

{
  "User-Agent": "MyApp/1.0",
  "X-Custom-Header": "custom-value"
}

This is useful for:

Sending a different User-Agent to providers than what the client sends
Adding provider-specific headers required by certain APIs
Overriding default headers for compatibility

Protocol Conversion

MRouter supports automatic bidirectional protocol conversion between Anthropic and OpenAI formats. This allows you to use any client (e.g., Claude Code sending Anthropic format) with any provider (e.g., DeepSeek expecting OpenAI format) — the proxy handles the translation transparently.

How It Works

Client format is auto-detected from the request path: /v1/messages → Anthropic, /v1/chat/completions → OpenAI
Provider format is determined by the provider's API Format setting
When the client format differs from the provider format, MRouter automatically converts requests, responses, and SSE streaming events

Configuration

Set the API Format field when adding/editing a provider:

| API Format | Behavior | |---|---| | Auto (default) | Passthrough — no conversion, requests forwarded as-is | | Anthropic | Provider expects Anthropic protocol. If client sends OpenAI format, auto-convert | | OpenAI | Provider expects OpenAI protocol. If client sends Anthropic format, auto-convert |

Example: Claude Code → DeepSeek

Claude Code sends Anthropic format (/v1/messages), but DeepSeek uses OpenAI format:

Add DeepSeek provider with API Format = OpenAI
Claude Code sends request to http://localhost:4444/v1/messages
MRouter detects mismatch (client=Anthropic, provider=OpenAI)
Request is converted: Anthropic → OpenAI format, path → /v1/chat/completions
Response/SSE stream is converted back: OpenAI → Anthropic format

What Gets Converted

Request: system message, message content format, max_tokens, stop sequences, path
Response: content blocks, finish reason, usage stats
SSE Streaming: Anthropic events (message_start, content_block_delta, message_delta, message_stop) ↔ OpenAI chunks (chat.completion.chunk, [DONE])

Provider Import/Export

Export and import provider configurations for easy migration between machines or backup.

Export (Shift+E): Saves all providers to ~/.mrouter/providers.json
Import (Shift+I): Load

Mrouter

Install / Use

README

MRouter - LLM Gateway Router

Why MRouter?

Key Use Cases

Core Features

Quick Start

Installation

Basic Usage

TUI Keyboard Shortcuts

Global

Providers Tab

Proxy Tab

Logs Tab

Stats Tab

Configuration

Supported Providers

Proxy Mode

Running as a Background Service

CLI Tool Configuration Examples

Custom Headers Per Provider

Protocol Conversion

How It Works

Configuration

Example: Claude Code → DeepSeek

What Gets Converted

Provider Import/Export