Dynamically convert OpenAPI specs into AI agent tools at runtime—no code generation required.
This monorepo contains four packages:
| Package | Description | Install |
|---|---|---|
| @spec2tools/core | Core utilities for OpenAPI parsing and authentication | npm install @spec2tools/core |
| @spec2tools/sdk | Create AI SDK tools from OpenAPI specifications | npm install @spec2tools/sdk |
| @spec2tools/cli | CLI for interacting with OpenAPI-based AI tools | npx @spec2tools/cli |
| @spec2tools/stdio-mcp | MCP server exposing OpenAPI endpoints as tools via stdio | npx @spec2tools/stdio-mcp |
import { createTools } from '@spec2tools/sdk';
import { generateText, stepCountIs } from 'ai';
import { openai } from '@ai-sdk/openai';
const tools = await createTools({ spec: './openapi.yaml' });
const result = await generateText({
model: openai('gpt-4o'),
tools,
prompt: 'List all users',
stopWhen: stepCountIs(3)
});
console.log(result.text);# Set your OpenAI API key
export OPENAI_API_KEY=your-api-key
# Start the agent with an OpenAPI spec
npx @spec2tools/cli start --spec https://api.example.com/openapi.json# Add to Claude Code (or any MCP client)
claude mcp add --transport stdio my-api \
-- npx @spec2tools/stdio-mcp ./path/to/openapi.yamlCollapse all API endpoints into just 2 tools (search + execute), cutting input token usage by ~99.9%:
# MCP server with code mode
claude mcp add --transport stdio my-api \
-- npx @spec2tools/stdio-mcp ./openapi.yaml --code-mode// SDK with code mode
const tools = await createTools({
spec: './openapi.yaml',
codeMode: true,
});You can also convert any existing AI SDK tools to code mode:
import { toCodeModeTools } from '@spec2tools/core';
const codeModeTools = toCodeModeTools(existingTools);You have a server. You have an OpenAPI spec. Now, with all the MCP hype, you're told you need to build another server—an MCP server—just so AI agents can call your API. You look it up: define tools, maintain handlers, keep it in sync with your actual API. But wait... you already described your API in OpenAPI. Why do we need to describe it to the agent twice?
Spec2Tools is the alternative. It reads your OpenAPI spec and exposes endpoints as tools—no code generation, no maintenance burden. Use it directly with the AI SDK, or spin up an MCP server instantly from your existing spec.
# Clone the repo
git clone https://github.com/CahidArda/spec2tools.git
cd spec2tools
# Install dependencies
pnpm install
# Build all packages
pnpm buildMIT