MCP Cost Optimizer MCP server

MCP Cost Optimizer

Local-first LLM/API cost analysis for AI agents, developers and teams. MIT licensed. Version 0.1.0 MVP.

Install in your AI client

Works with any MCP client over stdio; no account or API key needed for the local server.

Claude Code

claude mcp add cost-optimizer -- npx -y mcp-cost-optimizer

Codex CLI

codex mcp add cost-optimizer -- npx -y mcp-cost-optimizer

Claude Desktop, Cursor, Windsurf, Cline, Gemini CLI — add to the client's MCP config (claudedesktopconfig.json, ~/.cursor/mcp.json, ~/.codeium/windsurf/mcp_config.json, Cline MCP settings, ~/.gemini/settings.json):

{
  "mcpServers": {
    "cost-optimizer": {
      "command": "npx",
      "args": [
        "-y",
        "mcp-cost-optimizer"
      ]
    }
  }
}

VS Code / GitHub Copilot — .vscode/mcp.json:

{
  "servers": {
    "cost-optimizer": {
      "type": "stdio",
      "command": "npx",
      "args": [
        "-y",
        "mcp-cost-optimizer"
      ]
    }
  }
}

Zed — settings.json:

{
  "context_servers": {
    "cost-optimizer": {
      "command": "npx",
      "args": [
        "-y",
        "mcp-cost-optimizer"
      ]
    }
  }
}

Status

Core calculations and CLI tested. MCP SDK integration test is provided but could not be executed in the authoring environment because npm registry access is blocked. Run the full validation below before publishing. Package is not yet published to npm. No lockfile is included; generate and commit one after installing in your environment, then use npm ci in CI.

Quick start from source

Requires Node.js 20+.

npm install --ignore-scripts
npm run check
npm test
node test/mcp.integration.js
node src/index.js analyze examples/usage.jsonl examples/rates.json demo-budget

The CLI and core have no external dependencies. The MCP server uses the official TypeScript SDK v1 and Zod. No compilation is needed.

Configure your MCP host with an absolute path:

{
  "mcpServers": {
    "cost-optimizer": {
      "command": "node",
      "args": ["/absolute/path/mcp-cost-optimizer/src/index.js"]
    }
  }
}

Tools

Ask your agent: “Analyze these normalized usage records with my supplied rates. Identify unknown models and repeated requests, then compare a cheaper model. Explain assumptions and do not change production routing.”

Data contract

record: model, inputTokens, outputTokens, optional cachedInputTokens and requestHash. Input includes cached tokens; counts must be nonnegative safe integers. Usage must be normalized before import. OpenAI-style prompt_tokens and Anthropic cache fields are not accepted directly. Do not blindly map Anthropic input counts: sum ordinary input, cache-read and cache-write tokens first; cache-write pricing needs separate treatment and is unsupported in this MVP.

rates: three-letter currency, optional asOf, and models keyed by exact model name. Each model supplies inputPerMillion, outputPerMillion and optional cachedInputPerMillion. All models share the same supplied currency. Missing cache price falls back to normal input price. Real provider prices and currency conversion are deliberately not bundled; example rates are fictional. Unknown model costs are excluded and the report is marked incomplete; comparisons reject incomplete baselines.

Shortened. The full README is on GitHub.