apicmd

Turn any API into a CLI. Designed for LLM agents, works great for humans too.

Give it an OpenAPI spec URL — it generates a CLI with named operations, parameter validation, and --help that always shows the latest endpoints. No spec? Raw mode works with any HTTP API and learns endpoints from your usage history.

Install

# npm
npx apicmd

# bun
bunx apicmd

Quick Start

# Spec mode — point to an OpenAPI spec
apicmd init https://api.example.com/openapi.json --name myapi --auth 'Bearer $API_KEY'
apicmd myapi --help              # list all operations
apicmd myapi listUsers           # call by operation name

# Raw mode — any HTTP API, no spec needed
apicmd init https://api.example.com --name myapi --auth 'Bearer $API_KEY' --raw
apicmd myapi GET /users          # call any endpoint
apicmd myapi --help              # shows endpoints from your usage history

Setup

apicmd init <url> --name <name> [--auth <header>] [--ttl <hours>]   # Spec mode
apicmd init <url> --name <name> [--auth <header>] --raw             # Raw mode

| Flag | Description | |------|-------------| | <url> | OpenAPI spec URL (spec mode) or base URL (raw mode) | | --name | Short name for the API (used in all commands) | | --auth | Auth header, supports env vars: 'Bearer $MY_KEY' | | --ttl | Spec cache TTL in hours (default: 24) | | --raw | Raw mode — skip spec, use HTTP method + path |

Usage

Spec Mode

apicmd myapi --help                                    # list all operations with required params
apicmd myapi createIssue --help                        # show all params for an operation
apicmd myapi createIssue --title "Bug" --priority 2    # call an operation

Operations are auto-generated from the OpenAPI spec. The spec is cached and auto-refreshed based on the TTL (default 24 hours). --help always fetches the latest.

Raw Mode

apicmd myapi GET /api/users                                        # GET request
apicmd myapi POST /api/users --name "Alice" --email "[email protected]"      # POST with JSON body
apicmd myapi GET /api/users/{id} --id 123                          # path params via {placeholder}

How params are resolved:

• {key} in the path → replaced with --key value
• Remaining params on GET → query string
• Remaining params on POST/PUT/PATCH → JSON body
• Numbers and booleans are auto-coerced

Both Modes

Raw HTTP calls (apicmd myapi GET /path) work on any registered API, even spec mode ones. Use whichever is more convenient.

Auth

Auth supports env var references — the actual secret is never stored on disk:

apicmd init https://api.example.com/openapi.json --name myapi --auth 'Bearer $MY_API_KEY'

Literal keys also work:

apicmd init https://api.example.com/openapi.json --name myapi --auth 'Bearer sk-abc123'

Management

apicmd list          # show all registered APIs
apicmd --help        # show usage

Config

Stored in ~/.apicmd/<name>.json:

{
  "name": "myapi",
  "url": "https://api.example.com/openapi.json",
  "auth": "Bearer $MY_API_KEY"
}

That's it. The spec is cached after the first --help or operation call and auto-refreshed when stale.

For LLM Agents

Add this to your agent's global instructions (e.g. ~/.claude/CLAUDE.md):

Use `apicmd` CLI for API calls. Run `apicmd list` to see available APIs,
`apicmd <name> --help` for operations, `apicmd <name> <op> --help` for params.

The LLM runs --help once to discover all operations and required params, then calls them directly. No curl, no JSON construction, no auth headers.

License

MIT