High-performance browser automation bridge and multi-instance orchestrator with advanced stealth injection and real-time dashboard.
Attach Chrome
Use this guide when:
- Chrome already exists outside PinchTab
- you want the PinchTab server to register that browser as an instance
- you already have a browser-level DevTools WebSocket URL
Do not use this guide if your goal is simply:
- start a browser for your agent
- run the normal local PinchTab workflow
For that, use managed instances with pinchtab and POST /instances/start.
Launch vs attach
The mental model is:
launch = PinchTab starts and owns the browser
attach = PinchTab registers an already running browser
With attach:
- Chrome is started somewhere else
- PinchTab receives a
cdpUrl - the server registers that browser as an attached instance
What is implemented today
The current codebase implements:
POST /instances/attach— starts a childpinchtab bridge --cdp-attach ...process that wraps the external browser. The bridge speaks the normal PinchTab HTTP API; the orchestrator registers the bridge's HTTP URL (not the rawws://CDP URL) as the routable instance URL.POST /instances/attach-bridge— registers an already-running PinchTab bridge as an instance (unchanged).- attach policy in config under
security.attach - attached-instance metadata in
GET /instances
The attach request body is:
{
"name": "shared-chrome",
"cdpUrl": "ws://127.0.0.1:9222/devtools/browser/...",
"browser": "chrome"
}
browser (or its alias provider) is optional and takes a provider name
(chrome, cloak, ghost-chrome), not a target name. When browser.targets
is configured, the provider must have at least one configured target, and an
omitted value attaches with the default target's provider. If you pass both
browser and provider, they must agree. Without browser targets, the
provider defaults to chrome; use cloak for a CloakBrowser endpoint
(equivalent to --browser cloak on the CLI).
Accepted cdpUrl shapes:
- browser-level WebSocket URL:
ws://host:port/devtools/browser/<id> - HTTP DevTools origin:
http://host:port(resolved through/json/version) - HTTP
/json/versionURL
Page-level URLs (/devtools/page/...) are rejected.
There is currently no CLI attach command.
Step 1: enable attach policy
Attach is disabled unless you allow it in config.
Example:
{
"security": {
"attach": {
"enabled": true,
"allowHosts": ["127.0.0.1", "localhost", "::1"],
"allowSchemes": ["ws", "wss"],
"forwardProxyAuth": false
}
}
}
What this does:
- enables the attach endpoint
- restricts which hosts are accepted
- restricts which URL schemes are accepted
What this does not do:
- it does not start Chrome
- it does not define a global remote browser
- it does not replace managed instances
Step 2: start Chrome with remote debugging
Example:
google-chrome --remote-debugging-port=9222
# Or on some systems:
# chromium --remote-debugging-port=9222
This makes Chrome expose a browser-level DevTools endpoint.
Step 3: get the browser WebSocket URL
Query Chrome:
curl -s http://127.0.0.1:9222/json/version | jq .
# Response
{
"webSocketDebuggerUrl": "ws://127.0.0.1:9222/devtools/browser/abc123"
}
The value of webSocketDebuggerUrl is the cdpUrl you pass to PinchTab.
Step 4: attach it to PinchTab
curl -X POST http://localhost:9867/instances/attach \
-H "Authorization: Bearer $(pinchtab config token --stdout)" \
-H "Content-Type: application/json" \
-d '{
"name": "shared-chrome",
"cdpUrl": "ws://127.0.0.1:9222/devtools/browser/abc123",
"browser": "chrome"
}'
# Response (201, abridged)
{
"id": "inst_0a89a5bb",
"profileId": "prof_278be873",
"profileName": "shared-chrome",
"port": "9868",
"url": "http://127.0.0.1:9868",
"mode": "headed",
"headless": false,
"status": "running",
"attached": true,
"attachType": "cdp-bridge",
"cdpUrl": "ws://127.0.0.1:9222/devtools/browser/abc123",
"browser": "chrome"
}
Notes:
nameis optional; if omitted, the server generates one likeattached-...- the server validates the URL against
security.attach.allowHostsandsecurity.attach.allowSchemes(a rejected URL answers403) portandurlbelong to the child bridge PinchTab spawned from the instance port range, not to Chrome
Step 5: confirm it is registered
curl -s -H "Authorization: Bearer $(pinchtab config token --stdout)" http://localhost:9867/instances | jq .
# CLI Alternative
pinchtab instance list
An attached instance appears in the normal instance list with:
attached: trueattachType: "cdp-bridge"cdpUrl: ...status: "running"
Ownership and lifecycle
Attached instances are externally owned, but PinchTab still owns a bridge wrapper around them.
That means:
- PinchTab did not launch the browser
- PinchTab spawned a child
pinchtab bridge --cdp-attach ...process that wraps the external CDP endpoint and serves normal PinchTab routes - the external Chrome/CloakBrowser process remains outside PinchTab lifecycle ownership
In practical terms:
POST /instances/{id}/stopshuts down the child PinchTab bridge — the external Chrome process is left running- routes like
/tabs,/snapshot,/action,/screenshotgo to the child bridge, which talks CDP to the external browser through achromedp.NewRemoteAllocator
When attach makes sense
Use attach when:
- Chrome is managed by another system
- Chrome is already running in a separate service or container
- you want the server to know about an externally managed browser
- you want to keep browser ownership outside PinchTab
Security
Attach widens the trust boundary, so keep it locked down.
Recommended rules:
- leave attach disabled unless you need it
- keep
allowHostsnarrow - keep
allowSchemesnarrow - leave
forwardProxyAuthdisabled unless the attached browser process and CDP transport are trusted - set
PINCHTAB_TOKENwhen the server is reachable outside localhost - only attach to CDP endpoints you trust
If you set allowHosts to ["*"], PinchTab accepts any reachable attach host with an allowed scheme. That is a documented, non-default, security-reducing override: it removes host allowlisting entirely and should only be used on isolated, operator-controlled networks.
Also remember:
- Chrome DevTools gives powerful browser control
- a reachable CDP endpoint should be treated as sensitive infrastructure
If Chrome is remote, prefer a tunnel rather than exposing the debugging port broadly.
Operational model
The intended model is:
agent -> PinchTab server -> attached external Chrome
This is an expert path, not the default user path.
The default path remains:
pinchtab
then managed instance start via:
curl -X POST http://localhost:9867/instances/start \
-H "Authorization: Bearer $(pinchtab config token --stdout)" \
-H "Content-Type: application/json" \
-d '{"mode":"headless"}'
# CLI Alternative
pinchtab instance start
Related guides
- cloakbrowser.md — full CloakBrowser configuration and
the
--browser cloakattach variant - docker.md — running PinchTab (and the local CloakBrowser smoke image) in containers
- headed-mode.md — manual headed setup outside the bundled image
- security.md — attach policy details, IDPI, token handling