generated: '2026-09-13'
method: searched
source: https://www.withone.ai/docs/cli, https://www.withone.ai/docs/mcp,
  https://www.npmjs.com/package/@withone/cli, https://github.com/withoneai/cli
provider: IntegrationOS
providerId: integration-os
published: true
name: One CLI
binary: one
package: '@withone/cli'
version: 1.56.1
published_date: '2026-09-10'
repository: https://github.com/withoneai/cli
docs: https://www.withone.ai/docs/cli
install:
  - method: npm
    command: npm install -g @withone/cli
    source: https://www.withone.ai/docs/mcp
positioning: >-
  The CLI is the platform's primary surface, not an accessory to it — the provider's own framing is
  "One CLI to connect AI agents to every API on the internet" and the llms.txt quick start is a
  single line: `npm install -g @withone/cli && one init`. It is also the local MCP host: `one init`
  writes MCP server configuration for Claude Desktop, Cursor, Claude Code, Windsurf, Codex, Kiro or
  any MCP-compatible agent.
commands:
  - command: one init
    group: setup
    description: >-
      Prompts for an API key and configures the MCP server for the detected agent environment.
      Verbatim from https://www.withone.ai/docs/mcp.
    source: https://www.withone.ai/docs/mcp
  - command: one actions search <platform> "<query>"
    group: discovery
    description: >-
      Search the action catalogue for a platform. Verbatim from https://withone.ai/llms.txt —
      '"How to integrate Zoom/Stripe/Shopify with my AI agent" -- One CLI: one actions search
      <platform> "<query>"'.
    source: https://withone.ai/llms.txt
  - command: one --agent <command>
    group: global
    description: >-
      Agent mode — returns structured JSON instead of human output. The provider's own agent skill
      states that every third-party call an agent makes should carry this flag.
    source: https://github.com/withoneai/agent-skills/blob/main/integration-standards/SKILL.md
  - command: one --agent actions search <platform> "<what you want to do>"
    group: discovery
    description: Step 1 of the provider's documented three-step path — find the action.
    source: https://github.com/withoneai/agent-skills/blob/main/integration-standards/SKILL.md
  - command: one --agent actions knowledge <platform> <actionId>
    group: documentation
    description: >-
      Step 2 — read the action's documentation. The provider marks this REQUIRED before any execute.
    source: https://github.com/withoneai/agent-skills/blob/main/integration-standards/SKILL.md
  - command: one --agent actions execute <platform> <actionId> "$PLATFORM_KEY" [flags]
    group: execution
    description: >-
      Step 3 — execute. Flags: --path-vars for path variables, --query-params for query parameters,
      -d for the request body.
    source: https://github.com/withoneai/agent-skills/blob/main/integration-standards/SKILL.md
  - command: one --agent list
    group: discovery
    description: List your connections and their keys.
    source: https://github.com/withoneai/agent-skills/blob/main/integration-standards/SKILL.md
  - command: one add <platform>
    group: setup
    description: Connect a platform — e.g. `one add slack`.
    source: https://github.com/withoneai/agent-skills/blob/main/README.md
command_surface_note: >-
  Every invocation above is quoted verbatim from provider text — the MCP docs, llms.txt, or the
  provider's own published agent skill at https://github.com/withoneai/agent-skills. The full command
  tree is rendered client-side on https://www.withone.ai/docs/cli and is NOT enumerated here rather
  than guessed at; `one --help` on an installed binary is the authority. The CLI's functional areas,
  per the docs and llms.txt, are search, discover, authenticate and execute across 789 platforms,
  plus workflow (Flow) creation and structured JSON output.
agent_mode:
  flag: --agent
  description: >-
    Structured-JSON output mode intended for agents. The provider's published integration-standards
    skill treats `one --agent` as the ONLY sanctioned way for an agent to touch a third-party
    platform — "Never call a platform API directly with curl. Never install a platform SDK for a
    one-off call. Never paste a bearer token into a command."
  documented_path: search -> knowledge -> execute, in that order, with knowledge REQUIRED before execute
  source: https://github.com/withoneai/agent-skills/blob/main/integration-standards/SKILL.md
plan_availability:
  free: included
  starter: included
  pro: included
  enterprise: included
  source: https://www.withone.ai/pricing — "MCP & CLI" is a row available on every tier