Built as a Cloudflare Workers MCP server that counts tokens for AI model inputs. The implementation runs on Cloudflare's edge network and exposes both REST and MCP interfaces over JSON-RPC 2.0 via streamable HTTP transport. Reach for this when you need to estimate token usage before sending prompts to language models, especially if you're working with Claude or other token-limited APIs and want to avoid surprises on cost or context window limits. The remote endpoint is already live, so you can integrate it without deploying your own infrastructure. Useful for pre-flight checks in applications that construct dynamic prompts or need to stay within specific token budgets.