30-second demo
$ npx dynamic-openapi-cli -s https://petstore3.swagger.io/api/v3/openapi.json --help
Petstore - OpenAPI 3.0
Usage: dynamic-openapi-cli [command] [options]
Commands:
list-pets List all pets
get-pet-by-id Find pet by ID
create-pet Add a new pet to the store
update-pet Update an existing pet
delete-pet Deletes a pet
...
$ npx dynamic-openapi-cli -s ./petstore.yml get-pet-by-id 42
{
"id": 42,
"name": "Rex",
"status": "available"
}
$ dynamic-openapi-cli bundle -s ./petstore.yml --name petstore --out ./petstore
bundled "petstore" v1.0.0 → ./petstore (7.3 KB, 18 operations)
$ ./petstore install
petstore install: symlinked /home/ff/.local/bin/petstore → /home/ff/work/petstore
$ petstore list-pets --status=available --limit=5
$ petstore update
petstore update: spec changed (md5 bb864f70 → e5789514), … 1.0.0 → 1.1.0.
That's the whole product.
Quick start
npx dynamic-openapi-cli -s ./spec.yaml <command>
npm install -g dynamic-openapi-cli
dynamic-openapi-cli -s ./spec.yaml <command>
Three ways to use it, depending on what you need:
Mode 1 — ad-hoc (dev, scripts, CI)
dynamic-openapi-cli -s https://api.example.com/openapi.json \
list-pets --limit=20 --status=available
Reads the spec fresh every run. Good for scripting, first exploration, or CI jobs where the spec is always fetched from source.
Mode 2 — bundled (distribute a CLI)
dynamic-openapi-cli bundle \
-s ./openapi.yml \
--name my-api \
--out ./my-api
chmod +x ./my-api
./my-api install
my-api list-pets --limit=20
Packages the spec into a single-file bash executable. Commit it, symlink it, publish it to npm. Consumer only needs Node.js + npx (no install of dynamic-openapi-cli required — shim falls back to npx --yes).
Mode 3 — programmatic (TypeScript)
import { loadSpec, resolveSpec, buildCli, runCli } from 'dynamic-openapi-cli'
const spec = await resolveSpec(await loadSpec('./openapi.yml'))
const cli = buildCli({
spec,
name: 'my-api',
authConfig: { bearerToken: process.env.API_TOKEN! },
})
await runCli(cli, process.argv.slice(2))
Use the internals directly when you need a richer integration — custom auth, custom output formatting, or embedding the CLI in a larger tool.
Why this, and not curl / httpie / postman-cli?
| Reads OpenAPI spec | ✓ | — | — | ✓ |
| Operation names as subcommands | ✓ | — | — | ✓ |
| Path params as positional | ✓ | — | — | partial |
Query params as typed options with --help | ✓ | — | — | partial |
Enum values as choices | ✓ | — | — | — |
| Auto-auth from env | ✓ | — | — | ✓ |
| Works with file / URL / inline specs | ✓ | — | — | URL |
| Zero-install (via npx) | ✓ | ✓ | — | — |
| Bundle to single file | ✓ | — | — | — |
| Self-updating shim | ✓ | — | — | — |
| Pretty-print by default | ✓ | — | ✓ | ✓ |
| Works offline after bundle | ✓ | ✓ | ✓ | — |
curl -H "Authorization: Bearer $TOKEN" \
"https://api.example.com/pets?limit=20&status=available"
http GET api.example.com/pets Authorization:"Bearer $TOKEN" limit==20 status==available
petstore-cli list-pets --limit=20 --status=available
The CLI reads your OpenAPI spec and is the documentation: --help on any operation shows the path, method, parameters with their types and enum values, and where the request body fits.
How it works
┌─────────────────────────────┐
│ OpenAPI v3 spec │
│ (URL, file, or inline) │
└──────────────┬──────────────┘
│
loadSpec + dereference ($refs)
│
▼
┌─────────────────────────────┐
│ ParsedSpec │
│ operations, schemas, auth │
└──────────────┬──────────────┘
│
┌───────────┴───────────┐
▼ ▼
┌──────────────────┐ ┌────────────────────────┐
│ Dynamic CLI │ │ Bundle to bash shim │
│ (cli-args- │ │ ./my-cli │
│ parser) │ │ • SPEC_B64 (base64) │
│ │ │ • SPEC_MD5 │
│ each operation │ │ • SPEC_SOURCE │
│ → subcommand │ │ • install / update │
└────────┬─────────┘ │ • delegates to │
│ │ dynamic-openapi-cli │
▼ │ via npx │
HTTP request └────────────────────────┘
with auth + retry
│
▼
Pretty-printed JSON /
Raw body / -o file
One library, three entry points: the dynamic CLI for ad-hoc runs, the bundle command for packaging, and the programmatic TypeScript API for deeper integrations. All three share the same parser, HTTP client, and auth layer.
Command mapping
operationId: listPets | Command list-pets (kebab-cased) |
GET /pets/{petId} (no operationId) | Command derived from method + path |
summary / description | Command description shown in --help |
| Path params (in URL order) | Required positional args |
| Query / header / cookie params | --options with type coercion |
Parameter default | Option default shown in --help |
Parameter enum: [a, b, c] | choices — rejected values fail with error |
requestBody | --body '<json>' and --body-file <path> |
deprecated: true | [deprecated] prefix in description |
Example spec → CLI:
paths:
/pets/{petId}:
get:
operationId: getPetById
summary: Fetch a pet by its id
parameters:
- { name: petId, in: path, required: true, schema: { type: integer } }
- { name: include, in: query, schema: { type: string, enum: [owner, photos, history] } }
$ my-cli get-pet-by-id --help
my-cli get-pet-by-id
Fetch a pet by its id
Usage: my-cli get-pet-by-id <petId> [options]
Options:
--include <string> (choices: owner, photos, history)
$ my-cli get-pet-by-id 42 --include=photos
Response handling
application/json | Pretty-printed to stdout | Written verbatim | Saved as bytes |
Other text (text/*, XML, YAML) | Written verbatim | Written verbatim | Saved as bytes |
| Binary ≤ 256 KB | Inline JSON envelope with base64 data | same | Saved as raw bytes |
| Binary > 256 KB | Error: -o <file> required | same | Saved as raw bytes |
Every subcommand gets three global options for free:
-o, --output <file> Save response body to file
--raw Don't pretty-print JSON
-V, --verbose Print HTTP status + headers to stderr
Exit codes:
0 | Success (HTTP 2xx/3xx) |
1 | Network error, 5xx, or unexpected failure |
2 | Validation error or HTTP 4xx |
Bundle — the killer feature
One command turns any spec into a single-file bash executable that runs anywhere node runs.
dynamic-openapi-cli bundle \
-s ./openapi.yml \
--name petstore \
--out ./petstore
What's inside the shim?
- Full OpenAPI spec, base64-encoded and dereferenced (
$refs resolved).
- MD5 hash of the spec, baked into the file for change detection.
- Original source (URL or absolute path), so
update knows where to re-fetch.
- Runtime delegation to
dynamic-openapi-cli (global binary → npx --yes fallback → friendly error).
$ head -15 ./petstore
set -euo pipefail
CLI_NAME='petstore'
CLI_VERSION='1.0.0'
SPEC_SOURCE='https://petstore3.swagger.io/api/v3/openapi.json'
SPEC_SOURCE_KIND='url'
SPEC_MD5='bb864f7025e1408ccdc00f11f5c0e8bb'
SPEC_B64='eyJvcGVuYXBpIjoiMy4wLjMi...'
Subcommands every shim ships with
<any-operation> | Call the mapped API operation |
--help | Print help (top-level or per-command) |
--show-spec | Decode and print the embedded spec as JSON |
--spec-md5 | Print just the MD5 hash (scriptable) |
--spec <url|file> | Override the embedded spec at runtime (dev) |
update | Re-fetch the original spec and rewrite this file |
install | Symlink/copy into a PATH directory |
uninstall | Remove a previous install |
Self-update
The shim remembers where the spec came from. When the API evolves, update re-fetches, regenerates, and rewrites the file in place:
$ petstore --help
$ petstore --spec-md5
bb864f7025e1408ccdc00f11f5c0e8bb
$ petstore update
petstore update: fetching https://petstore3.swagger.io/api/v3/openapi.json ...
bundled "petstore" v1.1.0 → /home/ff/tmp/petstore.update.12345 (7.4 KB, 19 operations)
petstore update: spec changed (md5 bb864f70 → a12c9e31), /home/ff/bin/petstore 1.0.0 → 1.1.0.
$ petstore --help
The CLI version tracks the spec version. It's a snapshot of your API at a point in time, not a separate SemVer of the CLI tool. Override only if you actually need decoupled versioning:
petstore update
petstore update --app-version=2.0.0
petstore update --spec ./dev-spec.yml
Idempotent: running update when nothing changed is safe and prints spec unchanged (md5 …). Perfect for a cron job or a CI step.
0 3 * * * /usr/local/bin/petstore update >> /var/log/petstore-update.log 2>&1
If the bundle was built from an inline spec (raw JSON/YAML string, not a file or URL), update fails loudly — there's nothing to re-fetch. Re-run dynamic-openapi-cli bundle manually in that case.
Making the bundled CLI globally available
Three options, from "one command" to "npm ecosystem":
1. Built-in install (recommended)
./petstore install
./petstore install --dir /usr/local/bin
./petstore install --dir /usr/local/bin --copy
./petstore install --force
./petstore uninstall
Defaults to $XDG_BIN_HOME or $HOME/.local/bin — no sudo needed. If the target directory isn't on your PATH, the command prints the exact line to add to your shell rc:
petstore install: warning — /home/ff/.local/bin is not on your PATH yet.
Add this line to your shell rc (~/.bashrc, ~/.zshrc, or equivalent):
export PATH="/home/ff/.local/bin:$PATH"
Why symlink by default: running update in the original location transparently refreshes the globally-linked binary. Copy mode (--copy) is useful when you want the installed version frozen against further updates.
2. Manual move/symlink
Classic UNIX:
mv ./petstore /usr/local/bin/
ln -s "$PWD/petstore" /usr/local/bin/petstore
3. Publish as an npm package
Wrap the shim in a tiny package.json and publish — users then install with npm install -g <your-cli>:
{
"name": "petstore-cli",
"version": "1.0.0",
"bin": { "petstore-cli": "./petstore" },
"files": ["petstore"]
}
npm publish
npm install -g petstore-cli
petstore-cli list-pets
Good fit when you already have CI/CD publishing to npm and want your team's CLI to follow the same distribution path as other internal tools. Pair with a bundle step in CI to keep the shim up to date with the latest spec.
Authentication
Same variables and resolution order as dynamic-openapi-mcp — programmatic config → per-scheme env var → global env var.
| Bearer | OPENAPI_AUTH_TOKEN or OPENAPI_AUTH_<SCHEME>_TOKEN | auth.bearerToken |
| API Key (header/query/cookie) | OPENAPI_API_KEY or OPENAPI_AUTH_<SCHEME>_KEY | auth.apiKey |
| Basic | OPENAPI_AUTH_<SCHEME>_TOKEN as user:pass | auth.basicAuth |
| OAuth2 client credentials | — | auth.oauth2 (auto-refresh) |
| Custom token exchange | — | auth.tokenExchange (auto-refresh) |
| Fully custom | — | auth.custom (callback) |
Per-scheme env vars are derived from the securitySchemes name in your spec:
components:
securitySchemes:
bearerAuth:
type: http
scheme: bearer
maps to OPENAPI_AUTH_BEARERAUTH_TOKEN. Useful when a single CLI needs to talk to multiple auth realms.
OPENAPI_AUTH_TOKEN=sk-123 petstore list-pets
export OPENAPI_AUTH_TOKEN=$(vault read -field=token secret/petstore)
petstore list-pets
petstore create-pet --body-file=new-pet.json
CLI reference
dynamic-openapi-cli [bootstrap flags] <command> [...args]
Bootstrap flags (before the command):
-s, --source <url|file> OpenAPI spec URL, file path, or inline JSON/YAML
--spec <url|file> Alias for --source (used by bundled shims)
-b, --base-url <url> Override the base URL from the spec
--server-index <n> Pick the Nth server entry from the spec (default: 0)
--name <string> Display name in help (for bundled CLIs)
--app-version <string> Display version in help (for bundled CLIs)
--include-tag <name> Only expose operations with this tag (repeatable, comma-separated)
--exclude-tag <name> Hide operations with this tag (repeatable, comma-separated)
--include-operation <id> Only expose these operationIds (repeatable, comma-separated)
--exclude-operation <id> Hide these operationIds (repeatable, comma-separated)
--self-version Print dynamic-openapi-cli's own version
-h, --help Show help (global or per-command)
Global options (after the command):
-o, --output <file> Save response body to file
--raw Skip pretty-printing
-V, --verbose Print HTTP status + headers to stderr
Built-in subcommands (no spec required):
bundle Package a spec into a standalone bash CLI
(run "bundle --help" for details)
Environment
OPENAPI_SOURCE | Spec URL or file path (alternative to -s) |
OPENAPI_BASE_URL | Override the base URL |
OPENAPI_SERVER_INDEX | Select server entry (0-based) |
OPENAPI_AUTH_TOKEN | Global bearer token |
OPENAPI_API_KEY | Global API key |
OPENAPI_AUTH_<SCHEME>_TOKEN | Per-scheme bearer/basic token |
OPENAPI_AUTH_<SCHEME>_KEY | Per-scheme API key |
Filtering operations
Not every endpoint should show up as a subcommand. Two ways to trim:
Flags (and programmatic filters)
dynamic-openapi-cli -s ./spec.yaml --include-tag pets list-pets
dynamic-openapi-cli -s ./spec.yaml --exclude-tag admin
dynamic-openapi-cli -s ./spec.yaml --include-operation listPets,getPetById
dynamic-openapi-cli -s ./spec.yaml --include-tag pets --exclude-operation deletePet
Programmatic equivalent via buildCli:
const cli = buildCli({
spec,
filters: {
tags: { include: ['pets'], exclude: ['admin'] },
operations: { include: ['healthCheck'], exclude: ['deletePet'] },
},
})
Precedence (first match wins): x-hidden → operations.exclude → operations.include → tags.exclude → includes as allowlist. operations.include escapes a matching tags.exclude, but operations.exclude wins over everything except x-hidden.
x-hidden vendor extension
Hide an endpoint at the spec level — no flags required, applies to every consumer:
paths:
/admin/reset:
post:
operationId: adminReset
x-hidden: true
Use it for endpoints that appear in the public spec but should never become CLI subcommands, MCP tools, or skill operations.
Programmatic API
import {
loadSpec,
resolveSpec,
buildCli,
runCli,
executeOperation,
resolveAuth,
resolveBaseUrl,
} from 'dynamic-openapi-cli'
Build a full CLI from a spec
import { loadSpec, resolveSpec, buildCli, runCli } from 'dynamic-openapi-cli'
const spec = await resolveSpec(await loadSpec('./openapi.yml'))
const cli = buildCli({
spec,
name: 'my-api',
version: '2.0.0',
authConfig: { bearerToken: process.env.MY_API_TOKEN! },
defaultHeaders: { 'X-Client-Id': 'my-team' },
fetchOptions: { retries: 5, timeout: 20_000 },
})
const exitCode = await runCli(cli, process.argv.slice(2))
process.exit(exitCode)
Call an operation directly (bypass the CLI layer)
import { loadSpec, resolveSpec, executeOperation, resolveBaseUrl, resolveAuth } from 'dynamic-openapi-cli'
const spec = await resolveSpec(await loadSpec('./openapi.yml'))
const baseUrl = resolveBaseUrl(spec)
const auth = resolveAuth({ bearerToken: 'sk-…' }, spec.securitySchemes)
const op = spec.operations.find(o => o.operationId === 'listPets')!
const { response } = await executeOperation(
op,
{ limit: 20, status: 'available' },
{ baseUrl, auth },
)
console.log(response.status, await response.json())
Generate a bash bundle from your own tooling
import { buildBundle } from 'dynamic-openapi-cli'
await buildBundle({
source: 'https://api.example.com/openapi.json',
name: 'example-cli',
out: './dist/example-cli',
appVersion: '3.0.0',
})
Useful in a CI step that regenerates the shim on every spec change.
Installation
npm install -g dynamic-openapi-cli
npx dynamic-openapi-cli ...
pnpm add -D dynamic-openapi-cli
Requires Node.js 18+. TypeScript types are shipped.
The family
Three complementary projects, one spec, three output surfaces — pick the one that fits the use case:
dynamic-openapi-mcp | Live MCP server (stdio) | Every tool call spins the server | You want real-time introspection, auto-refreshed OAuth tokens, typed tool I/O |
dynamic-openapi-cli | Bash CLI (optionally bundled) | Humans and scripts invoke it | You want a commit-friendly shim humans and CI can run |
dynamic-openapi-skill | Static SKILL.md | Claude loads it on demand | You want zero runtime, diff-friendly docs, and model-driven calls via curl / fetch |
All three share the same parser and auth layer. Switching between them is a matter of pointing them at the same spec.
Head-to-head with the MCP sibling
Same parser, same auth, different consumer:
| Consumer | AI agents (Claude, Cursor, Windsurf, …) | Humans at a terminal |
| Operations become | MCP tools | CLI subcommands |
| Schemas become | MCP resources (openapi://schemas/…) | — |
| Auth | Env vars + programmatic | Env vars + programmatic (identical) |
| Transport | stdio / MCP protocol | stdout / stderr / exit codes |
| Bundle | — | Single-file bash shim |
| Self-update | — | update subcommand in the shim |
Pick the MCP version when you want AI agents to call your API. Pick the CLI version when you want humans, scripts, or CI jobs to call it. Pick the skill version when you want Claude to learn the API from a static markdown file — zero runtime required.
Roadmap
Got an idea? Open an issue.
License
MIT