Усильте агента
Собственная команда издателя. Не эндпоинт PlugMCP.
Host a copy with our keys and logs.
npx -y @autono/[email protected]У Claude Code нет ссылки для установки локальной команды. Скопируйте команду ниже и выполните её в терминале.
claude mcp add autono-buttons -- npx -y @autono/buttons@0.147.0Выполните в терминале.
{ "mcpServers": { "autono-buttons": { "command": "npx", "args": [ "-y", "@autono/[email protected]" ] } }}claude_desktop_config.json
Удалённый URL или команда на вашем компьютере? Чем различаются два способа установки
Запустить на PlugMCP
A new server you create. It is not this listing's official endpoint. Use the publisher install when you want their URL and their credentials.
ЗапускОт издателя
n8n for agents. A CLI workflow engine where AI agents build and run their own automations.
Each button is a self-contained, reusable action. Create it once, press it forever. Buttons wraps code, APIs, and agent instructions into a single interface with typed args and structured output.
n8n for agents. A deterministic workflow engine where AI agents build and run their own automations.
Buttons ships as a single static binary for macOS and Linux (amd64 + arm64). Windows support is tracked in autonoco/autono#350.
curl -fsSL https://raw.githubusercontent.com/autonoco/buttons/main/install.sh | sh
Installs to /usr/local/bin by default (uses sudo if needed). The script verifies the SHA256 checksum of every download. Override with env vars:
curl -fsSL https://raw.githubusercontent.com/autonoco/buttons/main/install.sh | BUTTONS_VERSION=v0.70.0 sh
curl -fsSL https://raw.githubusercontent.com/autonoco/buttons/main/install.sh | BUTTONS_INSTALL_DIR=$HOME/.local/bin sh
brew install autonoco/tap/buttons
Tap is auto-updated by goreleaser on every release. brew upgrade buttons to update.
Homebrew installs shell completions automatically, so tab completion works without a separate setup step.
go install github.com/autonoco/buttons@latest
Installs to $(go env GOPATH)/bin (usually ~/go/bin).
npm install -g @autono/buttons
Thin JS shim that resolves the platform binary via optionalDependencies.
docker pull ghcr.io/autonoco/buttons:latest
docker run --rm -v $PWD/.buttons:/home/buttons/.buttons \
ghcr.io/autonoco/buttons:latest press weather --arg city=Miami
Multi-arch Alpine image, ~5 MB, published to GHCR. Mount a volume to persist state between invocations.
buttons status # show available CLI/content updates
buttons update # install available CLI/content updates
buttons update --json # structured output
buttons update checks GitHub Releases for the CLI and refreshes floating button dependencies from .buttons/buttons.json and .buttons/buttons-lock.json. Normal interactive commands can also update before they run: buttons-auto-update refreshes floating button dependencies and defaults on; cli-auto-update updates the CLI binary when the throttle allows and defaults off. Homebrew installs are auto-detected and left to Homebrew unless cli-auto-update is enabled. To opt in, run buttons config set cli-auto-update true or set BUTTONS_CLI_AUTO_UPDATE=1. Docker users re-pull the image.
buttons version
buttons --version
# Create a button from an API
buttons create weather --url 'https://wttr.in/{{city}}?format=j1' --arg city:string:required -d "get weather"
# See what it does
buttons weather
# Press it
buttons press weather --arg city=Miami
# See the history
buttons history weather
Buttons can wrap code, HTTP APIs, imported files, or prompt-only instructions. Use --prompt as a modifier on any runnable button to attach instructions for the consuming agent.
# Scaffold + edit (default). Creates main.sh with a placeholder, tells you where to edit it.
buttons create deploy --arg env:string:required
# Then edit ~/.buttons/buttons/deploy/main.sh and press:
buttons press deploy --arg env=staging
# Or skip the scaffold with --code for one-liners
buttons create greet --code 'echo "Hello, $BUTTONS_ARG_NAME!"' --arg name:string:required
# Python scaffold
buttons create transform --runtime python --arg input:string:required
# Node scaffold
buttons create parse --runtime node
# Or import an existing script file
buttons create etl --file ./scripts/etl.sh --arg source:string:required
# GET with URL templates
buttons create weather --url 'https://wttr.in/{{city}}?format=j1' --arg city:string:required
# POST with JSON body
buttons create notify --url https://hooks.slack.com/services/xxx \
--method POST \
--header "Content-Type: application/json" \
--body '{"text": "{{message}}"}' \
--arg message:string:required
# GraphQL
buttons create repos --url https://api.github.com/graphql \
--method POST \
--header "Authorization: Bearer {{token}}" \
--header "Content-Type: application/json" \
--body '{"query": "{ repository(owner: \"{{owner}}\", name: \"{{repo}}\") { stargazerCount } }"}' \
--arg token:string:required --arg owner:string:required --arg repo:string:required
Import an existing script. The file is copied into the button folder.
buttons create deploy -f ./scripts/deploy.sh --arg env:string:required
The --prompt flag attaches an instruction for the consuming agent. Use it standalone or combine it with code/API buttons.
# Standalone: just an instruction
buttons create deploy-checklist \
--prompt "Before deploying, verify: 1) All tests pass 2) Staging is green 3) Team notified"
# Code + prompt: run the command, agent knows what to do with the output
buttons create check-logs \
--code 'northflank logs --service api --env production --tail 100' \
--prompt "Summarize any errors or warnings from these logs"
# API + prompt: call the API, agent interprets the result
buttons create analyze-weather \
--url 'https://wttr.in/{{city}}?format=j1' \
--arg city:string:required \
--prompt "Extract temperature and conditions as a one-line summary"
When pressed, the result includes both the output and the prompt:
{
"ok": true,
"data": {
"status": "ok",
"stdout": "ERROR 2026-04-18 connection timeout to db-primary...",
"prompt": "Summarize any errors or warnings from these logs",
"button": "check-logs"
}
}
buttons press weather --arg city=Miami
buttons press weather --arg city=Miami --json
buttons press weather --arg city=Miami --dry-run # validate args + render command, don't execute
buttons press slow-task --timeout 120
# Define at creation
buttons create deploy --code '...' --arg env:string:required --arg verbose:bool:optional
# Pass at press time
buttons press deploy --arg env=production --arg verbose=true
Types: string, int, bool, enum. Enum declares a fixed set of valid values — the CLI rejects anything outside the set and the board renders a picker:
buttons create deploy --code '...' --arg env:enum:required:staging|prod|canary
Code buttons get args as BUTTONS_ARG_<NAME> env vars. API buttons use {{arg}} template substitution.
Drawers chain buttons into repeatable workflows. Use them when one action needs multiple steps, data from an earlier step, loops, waits, or a triggerable wrapper.
buttons drawer create standup-digest
buttons drawer standup-digest add github-activity granola-query slack-post
buttons drawer standup-digest connect github-activity to slack-post
buttons drawer standup-digest set slack-post.args.channel='#standup'
buttons drawer standup-digest press
Step outputs are available to later steps through ${step_id.output.field} refs when the upstream button prints JSON.
buttons drawer release-flow set publish.args.version='${build.output.version}'
Drawers also support sub-drawers, loops, switches, aggregates, waits, webhook triggers, validation summaries, error handlers, and their own run history.
Triggers are the planned unified surface for running buttons and drawers automatically:
buttons trigger webhook linkedin-sync /linkedin-sync
buttons trigger cron slack-sync "0 */2 * * *"
buttons trigger hook docs-sync --enable "file:docs/**/*.md"
Webhook, cron, and hook triggers all target a button or drawer. Button targets can compile down to hidden one-step drawer wrappers so the runtime keeps one workflow model internally.
Current implementation: webhook-triggered drawers are live today with:
buttons drawer linkedin-sync trigger webhook /linkedin-sync
buttons webhook listen
The unified buttons trigger ... command is the planned surface over the same drawer model.
Registry installs use a package-manager style manifest and lockfile:
buttons add @your-desk/hello # writes .buttons/buttons.json and installs latest
buttons add @your-desk/deploy@1 # pins an exact version
buttons install # materializes .buttons/buttons.json
buttons status # reports available CLI/content updates
buttons update # updates floating dependencies
The root manifest is committed:
{
"schema_version": 1,
"dependencies": {
"@your-desk/hello": "latest",
"@your-desk/deploy": "1"
}
}
"latest" floats and can move when buttons update, passive OTA, or a publish-triggered wake runs. Exact versions are pins and do not move unless you change the manifest with another buttons add @desk/name@version.
.buttons/buttons-lock.json stores the exact resolved version, content hash, installed local name, kind, and resolution time. Installed runtime specs live in .buttons/buttons/<name>/button.json; they do not need source, source_name, or content_hash for update resolution.
Install from the hosted registry:
BUTTONS_REGISTRY_URL=https://registry.example buttons add @your-desk/hello
BUTTONS_REGISTRY_URL=https://registry.example buttons add @your-desk/[email protected]
BUTTONS_REGISTRY_URL=https://registry.example buttons install
Version flow:
# Publisher creates and publishes immutable version 1.
buttons create hello --code 'echo hello'
BUTTONS_REGISTRY_URL=https://registry.example buttons publish @your-desk/hello
# Agent workspace tracks latest and installs that published version.
BUTTONS_REGISTRY_URL=https://registry.example buttons add @your-desk/hello
# Publisher changes the source button and publishes again. If version 1 already
# exists, publish bumps button.json to version 2 and retries automatically.
BUTTONS_REGISTRY_URL=https://registry.example buttons publish @your-desk/hello
# The agent workspace can see and apply the new version.
buttons status
buttons update
Required peer buttons from button.json requires are installed transitively from the registry. requires uses the same shape as the root manifest: scoped package names mapped to "latest" or an exact version.
Expose local buttons over HTTP with the same press contract as the CLI:
buttons serve
buttons serve --host 0.0.0.0 --api-key "$(openssl rand -hex 16)"
Endpoints include GET /api/buttons, GET /api/buttons/{name}, POST /api/buttons/{name}/press, and GET /api/buttons/{name}/runs. HTTP buttons are gated behind --allow-http-buttons.
buttons # interactive card-grid board in a TTY, plain table when piped
buttons <name> # full-screen detail page (args, last run, script); press `e` to edit
buttons logs <name> # live log viewer — follow mode with `f`, jump with `g`/`G`, `Esc` to exit
buttons list --json # machine-readable list
buttons summary --json # workspace snapshot for agents
On the board, pressing a button with required arguments opens an inline form instead of erroring out. Fill it in, hit Enter, the press fires.
Personal defaults live in ~/.buttons/settings.json. Manage them with buttons config:
buttons config # show current values
buttons config set default-timeout 600 # default for future `buttons create`
buttons config set theme amber # board TUI theme: default | paper | phosphor | amber
buttons config unset theme # revert to built-in default
Env var overrides: BUTTONS_HOME (relocate the data directory), BUTTONS_THEME (one-shot theme override — useful for A/B).
Batteries are key/value pairs injected into every button press as BUTTONS_BAT_<KEY>=<value>. Use them to keep API tokens out of button scripts.
buttons batteries set APIFY_TOKEN apify_api_xxx # local by default inside a project, global otherwise
buttons batteries set OPENAI_KEY sk-... --global # ~/.buttons/batteries.json
buttons batteries list
buttons batteries get APIFY_TOKEN
buttons batteries rm OLD_KEY
Local (.buttons/batteries.json) overrides global on key collisions. Keys must match [A-Z][A-Z0-9_]*.
Every press is recorded as a JSON file in the button's pressed/ folder.
buttons history
buttons history weather
buttons history --last 5
buttons history weather --json
buttons delete deploy
buttons delete deploy -F # skip confirmation
buttons delete deploy --json # JSON mode implies force
buttons rm deploy # rm is an alias for delete
Each button is a self-contained folder:
~/.buttons/buttons/weather/
button.json # spec (args, runtime, timeout)
main.sh # code file (.sh, .py, .js based on runtime)
AGENT.md # agent instruction/context
pressed/ # run history
2026-04-18T09-53-45.json
Override storage location with BUTTONS_HOME.
Every command supports --json. Piped output auto-detects non-TTY and outputs JSON automatically.
{"ok": true, "data": {"status": "ok", "stdout": "...", "prompt": "...", "button": "weather"}}
{"ok": false, "error": {"code": "MISSING_ARG", "message": "...", "hint": "...", "spec": [...]}}
Error codes: NOT_FOUND, MISSING_ARG, VALIDATION_ERROR, TIMEOUT, SCRIPT_ERROR, RUNTIME_MISSING, STORAGE_ERROR, NOT_APPLICABLE, INTERNAL_ERROR, NOT_IMPLEMENTED.
By default, buttons create <name> scaffolds a shell button with a placeholder main.sh the agent can edit. Shortcut flags below let you skip the placeholder.
context.WithTimeout (default 300s)See SECURITY.md for the full threat model and how to report vulnerabilities.
Full CLI reference, concepts, and changelog live in the docs/ directory (published via Mintlify).
Buttons is licensed under the Apache License, Version 2.0.
Copyright 2026 Darley Ventures LLC dba Autono.
n8n for agents. A deterministic workflow engine where AI agents build and run their own automations.
{ "mcpServers": { "autono-buttons": { "command": "npx", "args": [ "-y", "@autono/[email protected]" ] } }}Инструменты
Списка инструментов пока нет. У удалённых серверов он появляется после первой проверки состояния.
FAQ
Скопируйте на этой странице конфигурацию stdio для Cursor, Claude Desktop, Claude Code или VS Code. Она использует установку издателя, а не прокси PlugMCP.
В этой записи не указано обязательных переменных окружения.
@autono/buttons работает локально через stdio.
Последняя проверка: 2026-11-14. n8n for agents. A deterministic workflow engine where AI agents build and run their own automations.