MCP 服务器

AgentCost MCP Server

Provides real-time AI model pricing, cost estimation, and budget management tools to help agents understand and optimize their spending. It enables agents to compare costs across multiple providers and select the most cost-effective models for specific tasks.

README

🤖 AgentCost MCP Server

Cost awareness for AI agents. Know what you're spending before the invoice shows up.

An MCP (Model Context Protocol) server that gives any AI agent real-time access to model pricing, cost estimation, budget checking, and model comparison. Built by an agent, for agents.

Why?

AI agents are flying blind on costs. They pick models without knowing the price, run tasks without budget awareness, and generate surprise bills. AgentCost fixes this by giving agents the tools to understand and optimize their own spending.

Tools

Tool	Description
`estimate_cost`	Estimate cost for a model + token count before making the call
`compare_models`	Compare costs across models, get cheapest/best-value/best-quality picks
`check_budget`	Check if usage fits a daily budget, get smart switch suggestions
`find_cheapest`	Find cheapest model for a task (coding, reasoning, writing, etc.)
`list_models`	Browse all 20+ models across 7 providers with pricing
`get_model`	Deep-dive on a specific model with reference costs

Quick Start

Install

npm install -g agentcost-mcp

Use with Claude Desktop

Add to your claude_desktop_config.json:

{
  "mcpServers": {
    "agentcost": {
      "command": "agentcost-mcp"
    }
  }
}

Use with any MCP client

agentcost-mcp  # Runs on stdio

Example: Agent Self-Optimization

An agent can call these tools to make smarter decisions:

Agent: "I need to process 50 customer emails. Let me check the cost first."

→ estimate_cost(model_id="anthropic/claude-sonnet-4", input_tokens=2000, output_tokens=500)
→ Result: $0.0135 per email, $0.675 total

Agent: "That's reasonable. But let me see if there's something cheaper..."

→ compare_models(input_tokens=2000, output_tokens=500, task="classification", min_quality=70)
→ Recommendation: "GPT-4.1 Nano ($0.0006/email) for classification. 98% cheaper."

Agent: "Perfect. I'll use Nano for classification, Sonnet for the complex replies."