Apertis MCP Server

An MCP server for managing your Apertis AI API Gateway account. Query models, check quota, manage API keys, and get intelligent model recommendations — all from Claude Code, OpenClaw, or any MCP-compatible AI agent.

Quick Start

Installation

npm install @apertis/mcp-server

Environment Setup

Set your Apertis API key:

export APERTIS_API_KEY="sk-..."
export APERTIS_BASE_URL="https://api.apertis.ai"  # Optional, defaults to api.apertis.ai

Running

apertis-mcp

Or with npm:

npx @apertis/mcp-server

Claude Code Integration

Add this to your ~/.claude/settings.json:

{
  "mcpServers": {
    "apertis": {
      "command": "npx",
      "args": ["@apertis/mcp-server"],
      "env": {
        "APERTIS_API_KEY": "sk-your-key-here"
      }
    }
  }
}

OpenClaw Integration

Add this to your openclaw.json:

{
  "mcpServers": {
    "apertis": {
      "command": "npx",
      "args": ["@apertis/mcp-server"],
      "env": {
        "APERTIS_API_KEY": "sk-your-key-here"
      }
    }
  }
}

Available Tools

| Tool | Description | |------|-------------| | list_models | List all available models with optional filtering (free/paid, capability) | | get_model_info | Get detailed info about a specific model (pricing, context, provider) | | compare_models | Side-by-side comparison of 2-5 models | | check_quota | Check your account balance, subscription status, and remaining quota | | get_usage_stats | Get usage statistics by model and time period (today/week/month) | | list_api_keys | List all your API keys with status and quota (keys are masked) | | create_api_key | Create a new API key with optional quota limit | | suggest_model | Freeform keyword-based model search over the full catalog | | recommend_model | Get the curated Apertis pick for a task type (coding, long-context, fast-chat, reasoning, vision) with live pricing | | delegate | Delegate bulk grunt work (bulk file reads, boilerplate, summarization) to a cheap intern model. Files are read by the coworker, so their content never enters your context — keeps your Claude usage limit intact |

Available Resources

| Resource | Description | |----------|-------------| | apertis://account | Current account info, balance, and subscription status | | apertis://models | Full catalog of available models (cached 5 min) | | apertis://usage/today | Today's usage summary by model |

Examples

Check Your Quota

Claude: check my remaining quota
MCP: Use the check_quota tool
→ Shows your account balance, subscription plan, and quota

Find a Model for Your Task

Claude: what model should I use for coding on a tight budget?
MCP: Use recommend_model with task="coding" and budget="low"
→ Returns the curated pick (e.g. deepseek-v3) with live pricing and alternatives

For freeform keyword search over the full catalog (e.g. "something good at translation"), use suggest_model instead.

Compare Models

Claude: compare gpt-4o and claude-opus-4
MCP: Use compare_models with models=["gpt-4o", "claude-opus-4"]
→ Side-by-side pricing, context window, and capability comparison

Check Usage

Claude: how much have I used this week?
MCP: Use get_usage_stats with period="week"
→ Shows total tokens/cost and breakdown by model

Manage API Keys

Claude: list my API keys
MCP: Use list_api_keys
→ Shows all keys with status and quota (masked for security)

Claude: create a new API key called "my-app" with 1M token quota
MCP: Use create_api_key with name="my-app" and quota=1000000
→ Shows the new key once (save it immediately!)

Delegate Grunt Work to a Coworker

The delegate tool gives Claude Code a cheap "intern". Claude stays the manager on your Anthropic subscription and hands off high-volume, low-judgement work — bulk file reads, boilerplate, summarization — to a cheap model through Apertis. The coworker reads files itself, so their bulk content never enters Claude's context, and the work runs on Apertis credit rather than your Claude usage limit.

Claude: summarize the errors in build.log
MCP: Use delegate with instruction="Summarize the errors in <=10 bullets"
     and file_paths=["build.log"]
→ Returns a concise summary; the full log never enters Claude's context

delegate parameters: instruction (required), file_paths (optional — read locally by the coworker), content (optional inline text), model (optional — defaults to deepseek-v4-flash, override via APERTIS_COWORKER_MODEL).

To make Claude delegate automatically, paste the routing rules from coworker-rules.md into your project's CLAUDE.md.

Environment Variables

• APERTIS_API_KEY (required): Your Apertis API key (starts with sk-)
• APERTIS_BASE_URL (optional): API base URL, defaults to https://api.apertis.ai
• APERTIS_COWORKER_MODEL (optional): Intern model for the delegate tool, defaults to deepseek-v4-flash

Security Notes

• All API keys in tool output are masked (first 4 + last 4 characters)
• When creating a new key, the full key is shown only once — save it immediately
• The MCP server requires your API key in the environment — keep it secure
• Use Claude Code environment variable expansion for credentials

Pricing Information

All pricing is displayed in USD per 1M tokens (standard unit). The server shows:

• Input token price (cost per 1M input tokens)
• Output token price (cost per 1M output tokens)
• Optional cache read/write pricing (for supported models)

Development

# Install dependencies
npm install

# Build TypeScript
npm run build

# Run in development
npm run dev

# Run tests (builds first, then runs unit + integration tests)
npm test

License

MIT