withoneai/cli

One CLI – A command‑line bridge that lets AI agents talk to any web service

What it is – A Node‑based CLI (one) that gives an AI‑powered agent a single, uniform way to call the APIs of more than 600 SaaS platforms (Gmail, Slack, Stripe, Shopify, HubSpot, …). The tool handles authentication, OAuth flows, rate‑limiting and request formatting for you, so the agent only needs to know what it wants to do, not how to talk to each service.

How it works

1. Setup – one init stores your WithOne API key (or logs you in via a browser) and installs a small “MCP” server into the AI agents you use (Claude, Cursor, etc.).
2. Connect a platform – one add <platform> runs an OAuth flow once and creates a connection key that represents the authenticated credential.
3. Discover actions – one actions search <platform> "send email" returns a list of API actions (identified by an actionId) that match a natural‑language query.
4. Read docs – one actions knowledge <platform> <actionId> shows the exact request schema, required parameters and examples.
5. Execute – one actions execute <platform> <actionId> <connectionKey> -d '{…}' sends the request, with the CLI injecting the right token and normalising the response.
6. Compose flows – Using one flow create … you can chain multiple actions (e.g., look up a Stripe customer then email them) into reusable, condition‑aware workflows that can be run with a single command.
7. Memory store – one mem … runs a local PostgreSQL‑backed store that can hold user notes and data synced from the connected platforms, enabling fast hybrid full‑text + semantic search for the agent.
8. Sync – one sync pulls data from the platforms into the memory store so the agent can query it without hitting the external APIs each time.

Key commands

Command	Purpose
one init	Authenticate, install the MCP server, choose global or project‑scoped config
one add <platform>	Perform OAuth and create a connection key
one list	Show all active connections and their access policies
one actions search …	Find an API action by natural‑language description
one actions knowledge …	Get full documentation for a specific action
one actions execute …	Call the action with JSON data, path vars, query params, etc.
one flow create …	Define a multi‑step workflow (conditions, loops, parallel steps)
one flow execute …	Run a saved workflow
one mem …	Manage the local unified memory store (add, search, list, import, etc.)
one sync	Pull platform data into the memory store for instant queries
one cache …	Inspect or clear the CLI’s cache of action metadata

Why it matters for AI agents

• No API‑key juggling – The agent only needs the single connection key; the CLI injects OAuth tokens behind the scenes.
• Uniform interface – All platforms expose actions in the same shape (method, path, JSON schema), letting a single LLM prompt template work everywhere.
• Safety – Access policies can be scoped (full, method‑level, or explicit allow‑list) so the agent knows its permissions before trying a call.
• Composable workflows – Complex business logic (e.g., “when a new Stripe customer appears, send a welcome email”) can be expressed as a flow file that the agent can trigger programmatically.
• Local memory – Syncing data into a PostgreSQL store gives the agent fast, searchable context without repeated remote calls, and supports optional semantic search via OpenAI embeddings.

Installation

# Quick one‑off install
npx @withone/cli@latest init
# Or globally
npm install -g @withone/cli
one init   # then follow the interactive prompts

Node 18+ is required; one sync needs Node 20 because it uses a pre‑built SQLite binary.

Typical use‑case

one add gmail                     # OAuth once, get a connection key
one actions search gmail "send email" -t execute
# Suppose the action ID returned is 12345
one actions execute gmail 12345 <key> -d '{"to":"jane@example.com","subject":"Hi","body":"Hello"}'

In a few commands the agent has sent an authenticated email without ever handling OAuth tokens or learning Gmail’s REST spec.

---

Bottom line – One CLI is a practical tool that abstracts away the plumbing of hundreds of SaaS APIs, giving AI agents a clean, secure, and programmable way to act on real‑world services.