OnlyCLI

Name: OnlyCLI
Author: OnlyCLI

Turn any OpenAPI spec into a complete CLI. One command. One binary. 35x fewer tokens than MCP.

OnlyCLI terminal demo: generate a CLI from the GitHub API spec, then query repos, search, and list issues

Every endpoint in the GitHub REST API (1,107 operations) becomes a typed CLI command with flags, help text, and shell completion.

Install

go install github.com/onlycli/onlycli/cmd/onlycli@latest

Or download a pre-built binary from Releases.

Quick Start

# 1. Generate -- point at any OpenAPI 3.x spec (local file or URL)
onlycli generate \
  --spec https://raw.githubusercontent.com/swagger-api/swagger-petstore/master/src/main/resources/openapi.yaml \
  --name petstore --out ./petstore-cli

# 2. Build
cd petstore-cli && go mod tidy && go build -o petstore .

# 3. Use
./petstore pet find-pets-by-status --status available --format table
./petstore --help

That's it. Every operation in the spec is now a CLI command.

What You Get

Every generated CLI ships with:

Capability	How
Every endpoint as a command	Grouped by tag, named by operationId
All parameters as `--flags`	Path, query, header, and request body fields
7 output formats	`--format json\|yaml\|table\|csv\|pretty\|jsonl\|raw`
Field projection	`--transform "#.name"` via GJSON
Go template output	`--template '{{.login}}'`
Auto-pagination	`--page-limit N` auto-detects Link, cursor, offset, page-number
Streaming	`--stream` for SSE and NDJSON responses
File input	`--data @payload.json` or `--data @-` for stdin
OAuth2 authentication	Device flow (interactive) + client credentials (CI)
Multi-profile config	`config set staging.token xxx && --profile staging`
Retry with backoff	`--max-retries 3` for transient 429/5xx errors
Dry-run mode	`--dry-run` prints the request without sending
HTTP/2 + gzip	Modern transport, automatic content decompression
Shell completion	`completion bash\|zsh\|fish\|powershell` with enum hints
Cross-platform	Linux, macOS, Windows; amd64, arm64

Examples

Pre-generated CLIs from live public specs:

Example	Source Spec	Groups	Commands
petstore	Swagger Petstore	3	19
github	GitHub REST API	43	1,107

Regenerate them yourself:

make generate-example

Comparison

	OnlyCLI	curl	Restish	MCP tools
Typed commands per endpoint	Yes	No	Yes	Varies
Single distributable binary	Yes	Yes	Yes	No
Output formatting (table/yaml/csv)	Yes	No	No	No
Multi-scheme pagination	Yes	No	No	No
Streaming (SSE/NDJSON)	Yes	Manual	No	Varies
Works offline, no server process	Yes	Yes	Yes	No
Token cost for AI agents	Low	Low	Low	High

Detailed write-ups: vs Stainless · vs Restish · vs curl/HTTPie

For AI Agents: 35x Cheaper Than MCP

A GitHub MCP server loads 55,000 tokens into every prompt. Three services consume 72% of your context window on idle. At scale, that is $51,000/month in pure schema overhead.

OnlyCLI takes the opposite approach:

Generate a binary once from the spec
Agent calls ./cli --help to discover commands (~200 tokens)
Agent runs ./cli repos get --owner x --repo y and reads stdout

Stable command names, predictable flags, structured JSON output. No MCP server, no schema injection, no connection management. See the full cost analysis.

Development

git clone https://github.com/onlycli/onlycli.git && cd onlycli
make test          # unit + integration tests
make lint          # golangci-lint
make check         # fmt + vet + lint + test
make build         # build to bin/onlycli

See CONTRIBUTING.md for the full guide and ARCHITECTURE.md for codebase structure.

License

MIT