mcpusage

Analyzes MCP (Model Context Protocol) server tool definitions and calculates token consumption for tool advertisement payloads using a built-in or configurable tokenizer. The tool parses tool schemas (name, description, input_schema) and computes tokens consumed when these tools are advertised to LLM clients, enabling developers to understand the cost of exposing tool catalogs in MCP servers.

Provides a command-line interface for processing multiple tool schemas or MCP server configurations in batch, computing aggregate and per-tool token metrics. The CLI accepts file paths or stdin input, parses tool definitions, and outputs results in configurable formats (JSON, table, summary), enabling integration into shell scripts and CI/CD pipelines for automated token budget validation.

Abstracts tokenizer implementation to support multiple backend tokenizers (e.g., tiktoken for OpenAI, custom tokenizers for other LLM providers), allowing users to measure token consumption using the same tokenizer their target LLM uses. The tool accepts a tokenizer configuration parameter and applies it consistently across all tool schema analysis, ensuring token counts match production LLM behavior.

Decomposes token consumption across individual tool schema components (tool name, description, input_schema, required fields, type definitions) and reports token counts per component. This granular analysis helps developers identify which parts of tool definitions consume the most tokens and where optimization opportunities exist, using a component-aware parsing strategy.