llm-token-tracker

Instructions

Implementation Reference

• src/mcp-server.ts:178-204 (handler)

  The main handler function that executes the 'track_usage' tool logic. It tracks input/output tokens for a given provider/model/user using the TokenTracker, calculates session cost, and returns a formatted text response.

  private trackUsage(args: any) {
    const { provider, model, input_tokens, output_tokens, user_id = 'current-session' } = args;
    
    const trackingId = this.tracker.startTracking(user_id);
    this.tracker.endTracking(trackingId, {
      provider: provider as 'openai' | 'anthropic' | 'gemini',
      model,
      inputTokens: input_tokens,
      outputTokens: output_tokens,
      totalTokens: input_tokens + output_tokens
    });
  
    const usage = this.tracker.getUserUsage(user_id);
    const totalTokens = input_tokens + output_tokens;
    const cost = usage?.totalCost || 0;
    
    return {
      content: [
        {
          type: 'text',
          text: `✅ Tracked ${totalTokens.toLocaleString()} tokens for ${model}\n` +
                `💰 Session Cost: ${formatCost(cost)}\n` +
                `📊 Total: ${usage?.totalTokens.toLocaleString() || 0} tokens`
        }
      ]
    };
  }

• src/mcp-server.ts:53-79 (schema)

  Input schema definition for the 'track_usage' tool, specifying required properties like provider, model, input_tokens, output_tokens.

  inputSchema: {
    type: 'object',
    properties: {
      provider: {
        type: 'string',
        enum: ['openai', 'anthropic', 'gemini'],
        description: 'AI provider'
      },
      model: {
        type: 'string',
        description: 'Model name'
      },
      input_tokens: {
        type: 'number',
        description: 'Input tokens used'
      },
      output_tokens: {
        type: 'number',
        description: 'Output tokens used'
      },
      user_id: {
        type: 'string',
        description: 'Optional user ID'
      }
    },
    required: ['provider', 'model', 'input_tokens', 'output_tokens']
  }

• src/mcp-server.ts:50-80 (registration)

  Registers the 'track_usage' tool in the MCP server's listTools response, including name, description, and input schema.

  {
    name: 'track_usage',
    description: 'Track token usage for an AI API call',
    inputSchema: {
      type: 'object',
      properties: {
        provider: {
          type: 'string',
          enum: ['openai', 'anthropic', 'gemini'],
          description: 'AI provider'
        },
        model: {
          type: 'string',
          description: 'Model name'
        },
        input_tokens: {
          type: 'number',
          description: 'Input tokens used'
        },
        output_tokens: {
          type: 'number',
          description: 'Output tokens used'
        },
        user_id: {
          type: 'string',
          description: 'Optional user ID'
        }
      },
      required: ['provider', 'model', 'input_tokens', 'output_tokens']
    }
  },

• src/mcp-server.ts:160-161 (registration)

  Dispatches 'track_usage' tool calls to the trackUsage handler method in the CallToolRequestHandler switch statement.

  case 'track_usage':
    return this.trackUsage(request.params.arguments);