takopi-preview 0.2.10

Tailscale preview command plugin for Takopi

takopi-preview

Expose local dev servers over Tailscale Serve with /preview commands.

quickstart

1. install takopi + this plugin (same environment as your transport).

uv tool install -U takopi --with takopi-transport-slack --with takopi-preview

2. make sure tailscale is up and magicdns is enabled (tailscale up).
3. add minimal config to ~/.takopi/takopi.toml:

[plugins]
enabled = ["takopi-transport-slack", "takopi-preview"]

[plugins.preview]
provider = "tailscale"

4. in chat, pick a worktree context and start a preview:

/myapp @feat/login
/preview start <port> use pnpm dev -- --host 127.0.0.1 --port <port>

Open the returned URL, then stop when done:

/preview stop

commands

• /preview start [port] [instruction...]: start a preview for the current context
• /preview list: show active previews (url, port, uptime, context)
• /preview stop [id|port]: stop a preview (defaults to current context)
• /preview killall: stop all previews (restricted by allowlist)
• /preview help: usage help

optional config

[plugins.preview]
path_prefix = "/preview"
ttl_minutes = 120
tailscale_https_port = 443
allowed_user_ids = [123456789]
local_host = "127.0.0.1"
tailscale_bin = "tailscale"
start_port = 5173
start_instruction = "use pnpm dev -- --host 127.0.0.1 --port 5173"
dev_server_start_timeout_seconds = 600

[plugins.preview.projects.myapp]
path_prefix = "/preview"
start_port = 3000
start_instruction = "start web subrepo dev server only"

Notes:

• provider = "tailscale" uses tailnet-only URLs from tailscale serve.
• ttl_minutes = 0 disables expiration.
• empty allowed_user_ids means no allowlist enforcement.

dev server prompting

/preview start asks Takopi to ensure the dev server is running for the current worktree before enabling Tailscale Serve.

• if the target port is already listening, Takopi assumes it is the right server and leaves it running (no diagnostics).
• if the port is closed, Takopi finds the right dev command (README, AGENTS, package scripts) and starts it, preferring pnpm > bun > npm > yarn or uv > poetry > pip.
• Takopi starts the dev server in a detached/background session so it keeps running after the command finishes (nohup/setsid/disown, with logs redirected).
• the server should bind to local_host (default 127.0.0.1) and the requested port; /preview start waits up to dev_server_start_timeout_seconds for the port to open (default: 90s).

All text after /preview start is forwarded to Takopi. If start_port is not configured, the preview port must be the first argument. When start_port is set, /preview start uses it by default and any arguments are treated as instruction text.

You can include flags directly in the instruction:

/preview start <port> use pnpm dev --host 127.0.0.1 --port <port>

/preview stop and /preview killall ask Takopi to stop the dev server if it is still listening on the preview port.

common setups

vite / web apps

Allow tailnet hosts and bind to localhost:

server: {
  host: "127.0.0.1",
  port: 5173,
  allowedHosts: [".ts.net"],
},

Start the dev server (or rely on /preview start), then run /preview start <port>.

react native (metro)

Metro expects requests at the root path, so use path_prefix = "/" and a dedicated HTTPS port for the Metro port (example: 8081).

metro.config.js example:

const config = getDefaultConfig(__dirname);
const metroPort = Number(process.env.METRO_PORT || 8081);
config.server = {
  ...config.server,
  port: metroPort,
};
module.exports = config;

Start Metro bound to localhost:

METRO_PORT=8081 bun start:dev -- --host localhost --port 8081

Expose it over tailnet:

/preview start 8081

On devices, set the dev server host/port to HOST.TAILNET.ts.net:8081 in the React Native dev menu.

If your dev client cannot use HTTPS, skip takopi-preview and connect directly to the tailnet IP by running Metro with --host 0.0.0.0.

state and ttl

• tailscale: sessions are derived from tailscale serve status; no preview state file is written.
• tailscale: if the requested port is already served, takopi will attempt to disable the existing serve entry before starting a new preview.
• tailscale: set path_prefix = "/" to serve from the tailnet root. This avoids subpath issues with apps that assume /, but only one preview can be served at a time with the built-in config. Use path_prefix = "/preview" if you need multiple concurrent previews.
• tailscale: the default HTTPS port is 443, so previews map to https://host.ts.net/preview/<port> (or https://host.ts.net/ when path_prefix = "/"). Set tailscale_https_port = 0 to use the preview port (for https://host.ts.net:3000/). When using per-port HTTPS, start the dev server on 127.0.0.1 (no --host 0.0.0.0) so tailscale can bind the public port without conflicts.

ttl_minutes controls automatic expiration for previews started by this takopi process; expired sessions are cleaned up on the next command invocation. worktrees that are pruned or deleted are also cleaned up on the next command. takopi shutdown stops all previews.

errors

• missing tailscale: follow the install docs and run tailscale up.
• serve disabled: enable serve for your tailnet (Tailscale admin UI) if you see the "Serve is not enabled" error.
• preview already active: if a port is already served, takopi will stop the existing serve entry before re-enabling it.
• service not reachable: ensure your dev server is running and bound to local_host (default 127.0.0.1); rerun /preview start if needed.
• not in a worktree: include a branch (ex: /myapp @feat/foo) to create/use a worktree.

spec alignment

this implementation follows the webapp preview workflow spec:

• command surface: start/list/stop/killall/help
• config in [plugins.preview] with per-project overrides
• tailscale serve for tailnet-only preview urls
• tailscale serve registry
• ttl-based expiration (ttl_minutes)
• allowlist enforcement via allowed_user_ids

license

mit