Skip to content
ctok

MCP Server

The @ctok/mcp package implements the Model Context Protocol and exposes four tools that Claude Code (and any MCP-compatible agent) can call directly.

Setup

Add to ~/.claude.json (global) or .claude/settings.json (per-project):

{
  "mcpServers": {
    "ctok": {
      "command": "npx",
      "args": ["-y", "@ctok/mcp"]
    }
  }
}

Restart Claude Code. Open a new conversation and you’ll see the ctok tools available.

Tools

estimate

Estimate token usage, cost, and model recommendation for a prompt.

estimate(prompt, taskType?, pastedCode?, projectContext?)

Returns token ranges, USD cost, recommended model + effort, and reduction suggestions.

refine

Run the 7-pass prompt refiner.

refine(prompt, context?)

Returns the refined prompt, token savings, specificity score 0-100, and per-pass details.

recommend_model

Get the best model and effort level for a task.

recommend_model(prompt, taskType?)

Returns model ID, effort level, complexity breakdown, and reasoning.

scan_project

Scan a local project directory.

scan_project(path?, topHeavyCount?)

Returns total token count, breakdown by extension, top-heavy files, and excluded file counts.

Example usage in Claude Code

User: Before I send this refactor task, estimate the tokens.


Claude: [calls estimate with the task prompt]
        Input: ~1,240 tokens · Output: ~600 · Cost: $0.054 at Sonnet 4.6
        Recommendation: sonnet-4-6, medium effort
        No reduction suggestions - context looks lean.

Run the server directly

Terminal window
npx -y @ctok/mcp

The server speaks JSON-RPC 2.0 over stdio and exits when stdin closes.