Turn your OpenAPI document to an MCP server in ~1000 tokens and 3 tools
Cuts API schema context from 147% to 0.2% using just 3 MCP tools instead of 183.
Turn any file into a CLI from TypeScript/JavaScript exports
Turns TypeScript functions into npm CLIs so AI agents spend fewer tokens than MCP.
Developers building tools for AI coding agents (Claude Code, Cursor)
oclif · pkg · Model Context Protocol
It reads your TypeScript/JavaScript exported functions and generates a complete npm/npx CLI package in one command:
npx clifast your-file.ts
It parses types, JSDoc comments and function signatures to generate a --help command which can be used by LLMs to navigate the available input arguments and use your files or repositories effectively with less input tokens.
Multiple exports become subcommands. External imports are bundled. The output is a ready-to-publish npm package (which comes with the benefits of npx).
The goal is to reduce token usage by exposing files or entire folders as CLI commands that can be executed by Claude Code or Cloudflare's Codemode using a reduced amount of input tokens while decreasing the need to build and maintain complex MCP servers when not needed.
Repo: https://github.com/AlexandrosGounis/clifast
Can you please give me some feedback if you find this helpful?
Cuts API schema context from 147% to 0.2% using just 3 MCP tools instead of 183.
Cuts AI agent token costs 80% by wrapping MCP servers as native shell commands.
MCP query server cuts codebase context by 87% with zero dependencies, measured.
AST + embeddings for codebase search—but Sourcegraph Cody, Cursor, and Continue already solve this.
Cuts MCP prompt tokens 46% with single Go binary, no Docker or vector DB.
Local indexer with AST + impact graph replaces grepping and cloud RAG for code context.