<h1 align="center">chrome-devtools-axi</h1>
<p align="center">
<a href="https://www.npmjs.com/package/chrome-devtools-axi"><img alt="npm" src="https://img.shields.io/npm/v/chrome-devtools-axi?style=flat-square" /></a>
<a href="https://github.com/kunchenguid/chrome-devtools-axi/actions/workflows/ci.yml"><img alt="CI" src="https://img.shields.io/github/actions/workflow/status/kunchenguid/chrome-devtools-axi/ci.yml?style=flat-square&label=CI" /></a>
<a href="https://github.com/kunchenguid/chrome-devtools-axi/actions/workflows/release-please.yml"><img alt="Release" src="https://img.shields.io/github/actions/workflow/status/kunchenguid/chrome-devtools-axi/release-please.yml?style=flat-square&label=Release" /></a>
<a href="#"><img alt="Platform" src="https://img.shields.io/badge/platform-macOS%20%7C%20Linux%20%7C%20Windows-blue?style=flat-square" /></a>
<a href="https://x.com/kunchenguid"><img alt="X" src="https://img.shields.io/badge/X-@kunchenguid-black?style=flat-square" /></a>
<a href="https://discord.gg/Wsy2NpnZDu"><img alt="Discord" src="https://img.shields.io/discord/1439901831038763092?style=flat-square&label=discord" /></a>
</p>
<h3 align="center">The most agent-ergonomic browser automation</h3>
`chrome-devtools-axi` wraps [chrome-devtools-mcp](https://www.npmjs.com/package/chrome-devtools-mcp) with an [AXI](https://axi.md)-compliant CLI.
- **Token-efficient** โ TOON-encoded output cuts token usage ~40% vs raw JSON
- **Combined operations** โ one command navigates, captures, and suggests next steps
- **Contextual suggestions** โ every response includes actionable next-step hints
## Benchmarks
Agent ergonomics is measurable.
The [axi benchmark](https://axi.md) runs the same 14 real-world browsing tasks (Wikipedia research, GitHub navigation, multi-site comparison, and more) through 7 browser automation setups - 5 repeats each, with `claude-sonnet-4-6` as the agent and an LLM judge scoring task success.
chrome-devtools-axi posts the lowest input tokens, cost, duration, and turn count of all 7 conditions, with 100% task success:
| Condition | Avg Input Tokens | Avg Cost/Task | Avg Duration | Avg Turns | Success |
| ------------------------------------ | ---------------- | ------------- | ------------ | --------- | -------- |
| **chrome-devtools-axi** | **79,141** | **$0.074** | **21.5s** | **4.5** | **100%** |
| dev-browser | 82,532 | $0.078 | 28.6s | 4.9 | 99% |
| agent-browser (Vercel) | 93,074 | $0.088 | 24.6s | 4.8 | 99% |
| chrome-devtools-mcp + compressor CLI | 130,779 | $0.091 | 29.7s | 7.6 | 100% |
| chrome-devtools-mcp + ToolSearch | 133,712 | $0.096 | 29.4s | 7.5 | 99% |
| chrome-devtools-mcp (raw MCP) | 184,711 | $0.101 | 26.0s | 6.2 | 99% |
| chrome-devtools-mcp code execution | 129,606 | $0.120 | 36.2s | 6.4 | 100% |
Against raw chrome-devtools-mcp - the very server this CLI wraps - that is 57% fewer input tokens, 26% lower cost, and 27% fewer agent turns.
## Quick Start
Install the chrome-devtools-axi skill in the [Agent Skills](https://agentskills.io) format with [`npx skills`](https://github.com/vercel-labs/skills):
```sh
npx skills add kunchenguid/chrome-devtools-axi --skill chrome-devtools-axi -g
```
That is the entire setup - no npm install needed.
The skill teaches your agent to run chrome-devtools-axi through `npx -y chrome-devtools-axi`, so the CLI comes along on demand.
The skill is not a user-facing slash command (`user-invocable: false`).
Just ask for anything that needs a real browser - opening a page, clicking through a flow, extracting page content, debugging console or network, auditing performance - and the agent loads the skill on its own when it recognizes the task.
For ordinary web search, curl-able pages, or static extraction, the skill tells agents to skip Chrome and use simpler fetch/curl-style tooling.
The skill frontmatter also includes Hermes Agent metadata (`author` plus `metadata.hermes` tags/category) so Hermes can list it as a first-class browser automation skill; other harnesses ignore those extra fields.
`-g` installs the skill for all projects (`~/.claude/skills/`, for example); drop it to install for the current project only (`.claude/skills/`).
## What Agent Sees
```sh
$ chrome-devtools-axi open https://example.com
page: {title: "Example Domain", url: "https://example.com", refs: 1}
snapshot:
RootWebArea "Example Domain"
heading "Example Domain"
paragraph "This domain is for use in illustrative examples..."
uid=g1:1 link "More information..."
help[1]:
Run `chrome-devtools-axi click @g1:1` to click the "More information..." link
$ chrome-devtools-axi click @g1:1
page: {title: "IANA โ IANA-Managed Reserved Domains", refs: 12}
snapshot:
...
```
Refs in snapshot output carry a `g<N>:` generation prefix that bumps every time a new accessibility tree is captured. Pass refs back exactly as printed. UID actions also verify that the tracked page has not mutated since that snapshot; if freshness cannot be confirmed (including for a legacy untagged ref), they fail loudly with `STALE_REF` instead of silently no-op'ing, so the agent re-snapshots and retries.
The skill also instructs agents to verify state-changing actions with a fresh snapshot, `eval`, or screenshot before reporting success, because a current ref can still produce no visible page change.
## Other Ways to Install
The skill is the recommended path, but it is not the only one.
### Zero setup
chrome-devtools-axi is an AXI, so any capable agent can run the CLI directly with nothing installed at all.
Just tell your agent:
```
Execute `npx -y chrome-devtools-axi` to get browser automation tools.
```
### Session hook
Want ambient browser context - including the live page state of an active session - fed into every agent session instead of loading on demand?
Install the CLI globally and opt into the hook:
```sh
npm install -g chrome-devtools-axi
chrome-devtools-axi setup hooks
```
This installs a `SessionStart` hook for **Claude Code**, **Codex**, and **OpenCode** that surfaces the current browser session and usage guidance at the start of each session.
**Restart your agent session after running this** so the new hook takes effect.
Development entrypoints such as `pnpm run dev` and `bin/chrome-devtools-axi.ts` are guarded from accidental hook installation.
### From source
```sh
git clone https://github.com/kunchenguid/chrome-devtools-axi.git
cd chrome-devtools-axi
pnpm install --frozen-lockfile
pnpm run build
pnpm link
```
## How It Works
The bridge keeps one persistent MCP session across CLI invocations. With no
shared URL (or a blank one), standalone mode uses the local stdio process chain
below. See [Configuration](#configuration) for the two shared-service choices.
```
โโโโโโโโโโโโโโโโโโโโโโโโโ
โ chrome-devtools-axi โ CLI โ parse args, format output
โโโโโโโโโโโโฌโโโโโโโโโโโโโ
โ HTTP (localhost:9224)
โผ
โโโโโโโโโโโโโโโโโโโโโโโโโ
โ Bridge Server โ Persistent process, manages MCP session
โโโโโโโโโโโโฌโโโโโโโโโโโโโ
โ stdio
โผ
โโโโโโโโโโโโโโโโโโโโโโโโโ
โ chrome-devtools-mcp โ Headless Chrome via DevTools Protocol
โโโโโโโโโโโโโโโโโโโโโโโโโ
```
In URL-only shared mode, the bridge uses Streamable HTTP directly instead:
```
โโโโโโโโโโโโโโโโโโโโโโโโโ
โ chrome-devtools-axi โ CLI โ parse args, format output
โโโโโโโโโโโโฌโโโโโโโโโโโโโ
โ HTTP (localhost:9224)
โผ
โโโโโโโโโโโโโโโโโโโโโโโโโ
โ Bridge Server โ Persistent per-session MCP client
โโโโโโโโโโโโฌโโโโโโโโโโโโโ
โ Streamable HTTP
โผ
โโโโโโโโโโโโโโโโโโโโโโโโโ
โ Shared MCP service โ One remote MCP process + Chrome
โโโโโโโโโโโโโโโโโโโโโโโโโ
```
- **Persistent bridge** โ a detached process keeps the selected MCP session alive across commands, so Chrome doesn't restart every invocation
- **Auto-lifecycle** โ the bridge starts on first command, writes a PID file to `~/.chrome-devtools-axi/bridge.pid`, and is reused only after a deep health check (`/health?deep=1`, one CDP `list_pages`). A bridge whose browser has died is terminated and respawned. Startup waits until `CHROME_DEVTOOLS_AXI_BRIDGE_TIMEOUT_MS` (default 30s). On stop, the bridge kills its process group; teardown escalates SIGTERM to SIGKILL, and only when `ps` confirms that PID is a bridge, so stdio-launched chrome-devtools-mcp and Chrome children are reaped. A bridge also shuts itself down when its stdio chrome-devtools-mcp process exits, since it can serve nothing without it. Otherwise it runs until `stop` unless you opt into an [idle timeout](#idle-timeout).
- **Snapshot parsing** โ accessibility tree snapshots are extracted and analyzed for interactive elements (`uid=` refs)
- **TOON encoding** โ structured metadata uses [TOON format](https://www.npmjs.com/package/@toon-format/toon) for compact, token-efficient output
## CLI Reference
### Navigation
| Command | Description |
| ----------------- | -------------------------------------------- |
| `open <url>` | Navigate to URL and snapshot |
| `snapshot` | Capture current page state |
| `screenshot <p>` | Save a screenshot to a file |
| `scroll <dir>` | Scroll: up, down, top, bottom |
| `back` | Navigate back |
| `reload` | Reload the selected page and snapshot |
| `wait <ms\|text>` | Wait for time or text to appear |
| `eval <js>` | Evaluate a JavaScript expression or function |
| `run` | Execute a multi-step script from stdin |
`eval` wraps plain input as `() => (<expr>)` before sending it to DevTools. For multi-statement logic, pass an arrow function or `function`. No-arg IIFE form `(...)()` is accepted too and unwrapped automatically.
```sh
chrome-devtools-axi eval "document.title"
chrome-devtools-axi eval "() => { const rows = [...document.querySelectorAll('tr')]; return rows.map((row) => row.textContent) }"
```
### Interaction
| Command | Description |
| -------------------------- | ------------------------------ |
| `click @<uid>` | Click an element by ref |
| `fill @<uid> <text>` | Fill a form field |
| `type <text>` | Type text at current focus |
| `press <key>` | Press a keyboard key |
| `hover @<uid>` | Hover over an element |
| `drag @<from> @<to>` | Drag an element onto another |
| `fillform @<uid>=<val>...` | Fill multiple form fields |
| `dialog <accept\|dismiss>` | Handle a browser dialog |
| `upload @<uid> <path>` | Upload a file through an input |
`press` delivers keys to the page via CDP `Input.dispatchKeyEvent`, not as
browser commands. Browser shortcuts such as `Meta+R`, `Ctrl+R`, and `Meta+L`
do not trigger browser actions. To refresh the selected page, use
`chrome-devtools-axi reload` (or `chrome-devtools-axi eval '() => location.reload()'`).
Like other snapshot-producing commands, `reload` accepts `--full` for an
untruncated snapshot.
### Page Management
| Command | Description |
| ----------------- | --------------------------- |
| `pages` | List all open tabs |
| `newpage <url>` | Open a new tab |
| `selectpage <id>` | Switch to a tab by ID |
| `closepage <id>` | Close a tab by ID |
| `resize <w> <h>` | Resize the browser viewport |
By default, `newpage` and `selectpage` print the automatic accessibility tree
and snapshot-derived suggestions. Use `--no-snapshot` when navigating to a page
and continuing with `eval` or another chained command without needing element
UIDs. It suppresses only the printed tree and its derived guidance: capture
still runs, preserving page metadata and fresh element-reference generations.
Leave it off for UID-based `click`/`fill` workflows, or run `snapshot` explicitly.
The flag may appear before or after the URL/page ID and takes precedence over
`--full`.
```sh
chrome-devtools-axi newpage https://example.com --no-snapshot
chrome-devtools-axi selectpage 1 --no-snapshot
```
Tabs share one cookie jar unless `newpage` is given `--context <name>`. That opens the tab in a separate browser context (chrome-devtools-mcp's `isolatedContext`) with its own cookies and storage, so one browser can hold two sign-ins at once or a signed-out view beside a signed-in one. Later `newpage --context` calls with the same name join that context. Names are kept by the bridge, so each named session has its own, and after the bridge stops a name starts a fresh context. Stopping the bridge closes Chrome when AXI launched it; in a browser something else started (`CHROME_DEVTOOLS_AXI_BROWSER_URL` or `CHROME_DEVTOOLS_AXI_AUTO_CONNECT`), tabs already open in the old context stay open. Names use 1-64 characters from `[A-Za-z0-9._-]`, not starting with `-`.
```sh
chrome-devtools-axi newpage https://example.com --context signed-out
```
`resize` reads the page's viewport back and fails, naming the actual size, when it did not reach the requested size (an emulated viewport, page zoom or the window's minimum size can hold it); use `emulate --viewport` then.
Page-scoped tools require a selected page. `pages` only lists tabs and never changes the selection. `selectpage` sets it. `newpage` (and `open` when it has to create a tab) selects the tab chrome-devtools-mcp reports as selected in its own listing, which is the tab it just created, even when that URL is already open in another tab or changed while loading; a title continuation, no marked row, more than one `[selected]` line anywhere in the listing, a repeated `## Pages` section or page id, or chrome-devtools-mcp's note that it moved its selection off a closed tab leaves the selection unset, and the next page-scoped command fails until `selectpage`. The selected column in `pages` is AXI's selection, not the MCP `[selected]` marker. Closing the selected tab clears it. There is no automatic retarget onto another tab.
If the browser reconnects in process, every page id is reissued and the selection is dropped. A command that named a page fails instead of running against a new id. `open` then creates a new tab rather than restoring the previous one. If a selected tab is already gone, the command instead fails with a page-unavailable error and clears that selection.
### Emulation
| Command | Description |
| --------- | ------------------------------- |
| `emulate` | Emulate device/network/viewport |
### DevTools Debugging
| Command | Description |
| ------------------ | ------------------------------ |
| `console` | List console messages |
| `console-get <id>` | Get a specific console message |
| `network` | List network requests |
| `network-get [id]` | Get a specific network request |
For large request or response bodies, prefer `network-get <id> --response-file <path>` or `--request-file <path>` so the body goes to disk instead of flooding agent context.
### Performance
| Command | Description |
| --------------------------- | ----------------------------- |
| `lighthouse` | Run a Lighthouse audit |
| `perf-start` | Start a performance trace |
| `perf-stop` | Stop the performance trace |
| `perf-insight <set> <name>` | Analyze a performance insight |
| `heap <path>` | Capture a heap snapshot |
### Bridge
| Command | Description |
| ------------- | ----------------------------- |
| `start` | Start the bridge server |
| `stop` | Stop the bridge server |
| `setup hooks` | Install or repair agent hooks |
### Maintenance
| Command | Description |
| ---------------- | ------------------------------------------------------ |
| `update` | Upgrade the installed CLI to the latest npm version |
| `update --check` | Report current vs latest version without installing it |
Running with no command shows the CLI home view. By default, it prepends `bin`
and `description` metadata (see [Configuration](#configuration) to customize the
description), then includes the current snapshot when a browser session is active
or the no-session status/help block when one is not.
### Flags
`--help`, `-v`, `-V`, and `--version` are top-level options. All other flags
are command-specific; the CLI rejects a flag that is not listed by
`chrome-devtools-axi <command> --help`.
| Flag | Description |
| --------------------------- | ------------------------------------------- |
| `--help` | Show usage information |
| `-v`, `-V`, `--version` | Show the installed CLI version |
| `--check` | Check for available updates (update) |
| `--full` | Show complete output without truncation |
| `--no-snapshot` | Omit auto snapshot (newpage/selectpage) |
| `--background` | Open new page in background (newpage) |
| `--context <name>` | Open in a separate context (newpage) |
| `--timeout <ms>` | Navigation timeout (open/newpage) |
| `--uid @<uid>` | Target a specific element (screenshot) |
| `--full-page` | Capture entire scrollable page (screenshot) |
| `--format <fmt>` | Image format: png, jpeg, webp (screenshot) |
| `--viewport <spec>` | Viewport like "390x844x3,mobile" (emulate) |
| `--color-scheme <value>` | dark, light, or auto (emulate) |
| `--network <condition>` | Network throttle: Slow 3G, etc. (emulate) |
| `--cpu <rate>` | CPU throttling rate 1-20 (emulate) |
| `--geolocation <lat>x<lon>` | Set geolocation (emulate) |
| `--user-agent <string>` | Custom user agent (emulate) |
| `--type <type>` | Filter by type (console, network) |
| `--limit <n>` | Max items to return (console, network) |
| `--page <n>` | Pagination (console, network) |
| `--device <device>` | desktop or mobile (lighthouse) |
| `--mode <mode>` | navigation or snapshot (lighthouse) |
| `--output-dir <path>` | Directory for reports (lighthouse) |
| `--no-reload` | Skip page reload (perf-start) |
| `--no-auto-stop` | Disable auto-stop (perf-start) |
| `--file <path>` | Save trace data to file (perf-start/stop) |
| `--response-file <path>` | Save response body (network-get) |
| `--request-file <path>` | Save request body (network-get) |
Local output paths for `screenshot`, `heap`, `network-get --response-file`/`--request-file`, `lighthouse --output-dir`, and `perf-start`/`perf-stop --file` resolve against the directory where you invoke the CLI.
Saved-path output uses the resolved absolute path.
A denied or failed write exits non-zero with the tool error instead of printing success.
Snapshots truncate at about 16k characters unless `--full` is set. `eval` output keeps its head and tail when truncated.
Relative file paths for `upload` also resolve against the directory where you invoke the CLI, even if the bridge started in a different directory. Absolute upload paths pass through unchanged.
`console --type` accepts `log`, `debug`, `info`, `error`, `warn`, `dir`, `dirxml`, `table`, `trace`, `clear`, `startGroup`, `startGroupCollapsed`, `endGroup`, `assert`, `profile`, `profileEnd`, `count`, `timeEnd`, `verbose`, `issue`, and `all`.
`network --type` accepts `document`, `stylesheet`, `image`, `media`, `font`, `script`, `texttrack`, `xhr`, `fetch`, `prefetch`, `eventsource`, `websocket`, `manifest`, `signedexchange`, `ping`, `cspviolationreport`, `preflight`, `fedcm`, `other`, and `all`.
For both commands, `all` or an omitted `--type` returns every item.
## Configuration
Set `CHROME_DEVTOOLS_AXI_DESCRIPTION` to replace the home view's description with
text that fits your browser-tool routing rules. Leave it unset to keep the built-in
description, or set it to an empty string to omit the description line while
keeping the session status and help:
```sh
export CHROME_DEVTOOLS_AXI_DESCRIPTION="Use for public pages and local development."
# Or omit the description:
export CHROME_DEVTOOLS_AXI_DESCRIPTION=""
```
AXI honors chrome-devtools-mcpโs usage-statistics opt-out by forwarding `CHROME_DEVTOOLS_MCP_NO_USAGE_STATISTICS` and `CI` to the MCP server it launches.
The bridge server port defaults to `9224`. Override it with an environment variable:
```sh
export CHROME_DEVTOOLS_AXI_PORT=9225
```
To share one long-lived Chrome DevTools MCP service across AXI sessions on the
same host, set `CHROME_DEVTOOLS_AXI_MCP_SERVER_URL` to the service's Streamable
HTTP endpoint. The exact combination with `CHROME_DEVTOOLS_AXI_MCP_PATH` selects
how AXI reaches it:
1. **Direct (recommended, simplest):** the shared URL is nonblank and
`CHROME_DEVTOOLS_AXI_MCP_PATH` is absent or blank. AXI constructs a
`StreamableHTTPClientTransport` directly, starts no local MCP child, and
validates that the URL is an absolute `http://` or `https://` endpoint. The
shared service must already be running and reachable. This uses exactly one
MCP process for all named AXI bridges; each bridge still gets its own remote
MCP session/context and selected-page state.
```sh
MCP_BIN=/absolute/path/to/chrome-devtools-mcp/build/src/bin/chrome-devtools-mcp.js
node "$MCP_BIN" --http-port=9333 --isolated --headless \
--chrome-arg=--use-mock-keychain --chrome-arg=--password-store=basic &
unset CHROME_DEVTOOLS_AXI_MCP_PATH
export CHROME_DEVTOOLS_AXI_MCP_SERVER_URL=http://127.0.0.1:9333/mcp
CHROME_DEVTOOLS_AXI_SESSION=agent-a chrome-devtools-axi pages
CHROME_DEVTOOLS_AXI_SESSION=agent-b chrome-devtools-axi pages
```
2. **Stdio proxy:** the shared URL and `CHROME_DEVTOOLS_AXI_MCP_PATH` are both
nonblank. AXI checks that executable's `--help` advertises `--serverUrl`,
then starts it with only `--server-url=<URL>`. Use this when the selected
`chrome-devtools-mcp` build provides the verified proxy mode from
[ChromeDevTools/chrome-devtools-mcp#2733](https://github.com/ChromeDevTools/chrome-devtools-mcp/pull/2733).
The AXI bridge starts one proxy child per named bridge; manage the shared
service separately.
```sh
export CHROME_DEVTOOLS_AXI_MCP_PATH=/absolute/path/to/chrome-devtools-mcp/build/src/bin/chrome-devtools-mcp.js
node "$CHROME_DEVTOOLS_AXI_MCP_PATH" --http-port=9333 --isolated --headless \
--chrome-arg=--use-mock-keychain --chrome-arg=--password-store=basic &
export CHROME_DEVTOOLS_AXI_MCP_SERVER_URL=http://127.0.0.1:9333/mcp
CHROME_DEVTOOLS_AXI_SESSION=agent-a chrome-devtools-axi pages
CHROME_DEVTOOLS_AXI_SESSION=agent-b chrome-devtools-axi pages
```
If `CHROME_DEVTOOLS_AXI_MCP_SERVER_URL` is absent or blank, AXI keeps its
standalone stdio behavior: it starts the selected local
`chrome-devtools-mcp` process (in order: `CHROME_DEVTOOLS_AXI_MCP_PATH`, a
`chrome-devtools-mcp` executable on `PATH`, a detected global install, or
`npx -y chrome-devtools-mcp@latest`) and launches/attaches
Chrome according to the local settings below. A nonblank MCP_PATH without a
shared URL also remains this local stdio mode.
`CHROME_DEVTOOLS_AXI_MCP_SERVER_URL` takes precedence over AXI's local Chrome
launch and attach settings. Run the shared service on the same host and
filesystem as AXI, keeping the endpoint loopback-only, so saved artifact paths
refer to the same local files. Generic MCP stdio clients remain supported; the
proxy choice above only changes how AXI's bridge reaches the shared service.
Stop any existing bridges for those session names before switching between
these configurations: a running bridge retains its original transport
settings. `CHROME_DEVTOOLS_AXI_SESSION=<name> chrome-devtools-axi stop` stops
that session's bridge and any AXI-owned proxy child; manage the shared service's
lifecycle separately.
Connect to an existing Chrome instance instead of launching one:
```sh
export CHROME_DEVTOOLS_AXI_BROWSER_URL=http://127.0.0.1:9222
```
`CHROME_DEVTOOLS_AXI_BROWSER_URL` accepts both `http://` or `https://` URLs and `ws://` or `wss://` endpoints:
- `http(s)://` uses `--browserUrl` and fetches `/json/version` to discover the WebSocket URL.
- `ws(s)://` uses `--wsEndpoint` directly.
For authenticated `ws://` or `wss://` endpoints, pass JSON headers with `CHROME_DEVTOOLS_AXI_WS_HEADERS`:
```sh
export CHROME_DEVTOOLS_AXI_BROWSER_URL=wss://cluster.example/launch
export CHROME_DEVTOOLS_AXI_WS_HEADERS='{"Authorization":"Bearer token"}'
```
Pick which installed Chrome release channel to target with `CHROME_DEVTOOLS_AXI_CHANNEL` - `stable` (the default), `beta`, `canary`, or `dev`:
```sh
export CHROME_DEVTOOLS_AXI_AUTO_CONNECT=1
export CHROME_DEVTOOLS_AXI_CHANNEL=beta
```
This selects which Chrome `--autoConnect` attaches to, and which one is launched in the default and `CHROME_DEVTOOLS_AXI_USER_DATA_DIR` modes.
It is ignored when `CHROME_DEVTOOLS_AXI_BROWSER_URL` is set, since that connects to an explicit endpoint regardless of channel.
Launch a specific Chrome binary instead of the installed Chrome with `CHROME_DEVTOOLS_AXI_EXECUTABLE_PATH` (chrome-devtools-mcp's `--executablePath`).
The installer prints the executable path of the downloaded browser; set the variable to that printed path, which depends on your platform and the current stable version:
```sh
npx @puppeteer/browsers install chrome@stable --path ~/.cache/puppeteer
# macOS arm64 example; use the path the installer printed
export CHROME_DEVTOOLS_AXI_EXECUTABLE_PATH="$HOME/.cache/puppeteer/chrome/mac_arm-155.0.8059.39/chrome-mac-arm64/Google Chrome for Testing.app/Contents/MacOS/Google Chrome for Testing"
```
It applies only when this tool launches the browser (the default `--isolated` mode and `CHROME_DEVTOOLS_AXI_USER_DATA_DIR`) and is ignored in the `CHROME_DEVTOOLS_AXI_AUTO_CONNECT` and `CHROME_DEVTOOLS_AXI_BROWSER_URL` modes, where the browser is already running.
For launched browsers it takes precedence over `CHROME_DEVTOOLS_AXI_CHANNEL`, which is then not passed.
On macOS, prefer a [Chrome for Testing](https://developer.chrome.com/blog/chrome-for-testing) build: a headless copy of the installed Chrome registers with LaunchServices under the installed browser's bundle id, so links clicked in other apps can be delivered to a windowless automation instance instead of the user's browser. Chrome for Testing uses its own bundle id and can never be chosen.
### Keychain isolation
When chrome-devtools-axi launches Chrome itself - the default `--isolated` mode and `CHROME_DEVTOOLS_AXI_USER_DATA_DIR` - it always passes `--use-mock-keychain` and `--password-store=basic`.
An automation browser has no business reading, writing, or offering to reset your OS password store, so it is kept off it entirely.
Password autofill and saved-password access are therefore intentionally unavailable inside browsers this tool launches.
On macOS this also means the browser can never raise the system "Keychain Not Found ... Reset To Defaults" panel, which Chrome triggers when it tries to store its `Chrome Safe Storage` key and no default keychain can be resolved for the process.
Your own externally launched Chrome is unaffected: its saved passwords remain available and untouched because this tool does not read, write, move, or reset the login keychain or its `Chrome Safe Storage` item.
The isolation flags apply only to browsers this tool starts and are deliberately not sent in the `CHROME_DEVTOOLS_AXI_AUTO_CONNECT`, `CHROME_DEVTOOLS_AXI_BROWSER_URL`, and `wsEndpoint` modes, where the browser belongs to whoever launched it.
The shared MCP service is also externally launched; its operator owns Chrome's keychain policy, as illustrated by the service launch example above.
Run multiple isolated bridges at once with `CHROME_DEVTOOLS_AXI_SESSION` - one per agent session, worktree, or test worker:
```sh
CHROME_DEVTOOLS_AXI_SESSION=worker-1 chrome-devtools-axi open https://example.com
CHROME_DEVTOOLS_AXI_SESSION=worker-2 chrome-devtools-axi open https://example.org
```
Each session name gets its own bridge process, port (auto-derived from the name, or pinned with `CHROME_DEVTOOLS_AXI_PORT`), and on-disk state.
Outside shared-service mode, the default `--isolated` and `CHROME_DEVTOOLS_AXI_USER_DATA_DIR` launch modes each launch a Chrome per bridge, so concurrent sessions share neither browser state nor each other's stale-ref tracking.
Sessions that attach to the same external browser - multiple `CHROME_DEVTOOLS_AXI_AUTO_CONNECT=1` sessions on one running Chrome, or the same `CHROME_DEVTOOLS_AXI_BROWSER_URL`/`wsEndpoint` - drive that shared browser and are isolated only at the bridge level, where the per-session generation counter does not prevent cross-talk.
A session name does not choose a connection mode or profile; for a locally launched persistent browser, give each session its own `CHROME_DEVTOOLS_AXI_USER_DATA_DIR`.
The default (unset) session keeps port 9224 and the legacy state paths below.
Do not export `CHROME_DEVTOOLS_AXI_PORT` globally when running concurrent sessions: it overrides the per-session derived port and forces every session onto the same port, so the second session fails to start - its bridge cannot bind the already-taken port, and the first session's bridge is rejected as a mismatch rather than silently shared.
Rely on the per-session default ports instead, or set `CHROME_DEVTOOLS_AXI_PORT` only inline per command.
State is stored in `~/.chrome-devtools-axi/` (named sessions nest under `sessions/<name>/`):
| File | Purpose |
| --------------------- | ------------------------------------- |
| `bridge.pid` | PID and port of the running bridge |
| `snapshot-generation` | Counter used to detect stale uid refs |
### Navigation timeout
chrome-devtools-mcp gives a navigation 10 seconds by default, so a heavy page can fail with `Navigation timeout of 10000 ms exceeded`. Opt into a longer wait with `--timeout <ms>` on `open` or `newpage`, or set a default once:
```sh
chrome-devtools-axi newpage https://example.com/report --timeout 60000
export CHROME_DEVTOOLS_AXI_NAVIGATION_TIMEOUT_MS=60000
```
The flag wins over the environment variable. Either must be a positive integer (at most 1000000000); anything else is a usage error. Without either, nothing is sent and the 10-second default applies. The bridge raises its own MCP request deadline above the navigation timeout, so a long navigation is not cut off first.
### Idle timeout
A bridge runs until `chrome-devtools-axi stop` by default, because it is detached from the shell or agent session that started it.
To have bridges clean up after sessions that never run `stop`, opt into an idle timeout in milliseconds:
```sh
export CHROME_DEVTOOLS_AXI_IDLE_TIMEOUT_MS=1800000 # 30 minutes
```
A bridge started with it set shuts itself down, together with the chrome-devtools-mcp process and any browser it launched, once that long passes with no command and no request in flight.
Every command that talks to the browser counts as activity except the home view: bare `chrome-devtools-axi`, which is also the page summary `setup hooks` shows at agent session start, does not renew the timeout, so starting new agent sessions does not keep an abandoned bridge alive.
The next command after a shutdown starts a fresh bridge, so pages and state of a browser this tool launched are lost, as after `stop`.
Unset, blank, or `0` keeps the default; values below `1000` are raised to `1000`, and a value that is not a whole number of milliseconds is ignored.
A running bridge keeps the setting it started with.
## Development
```sh
pnpm run build # Compile TypeScript to dist/
pnpm run build:skill # Regenerate the minimal skills/chrome-devtools-axi/SKILL.md stub from src/skill.ts
pnpm run dev # Run CLI directly with tsx
pnpm test # Run tests with vitest
pnpm run test:watch # Run tests in watch mode
```
The committed `skills/chrome-devtools-axi/SKILL.md` is a minimal discovery stub generated by `pnpm run build:skill`; it points to the CLI for current guidance, and `pnpm test` fails if it drifts from `createSkillMarkdown()`.
The generated skill and its frontmatter are owned by `src/skill.ts`; update the generator instead of hand-editing the committed `SKILL.md`.
The npm package includes `skills/chrome-devtools-axi/`, so published releases ship the same installable Agent Skill documented in Quick Start.
Prettier intentionally ignores generator-owned files listed in `.prettierignore`; use their generator checks instead of formatting them by hand.