Allows observe-only access to Brave browser tabs via the pagouse Chromium extension, including listing open tabs and capturing accessibility-tree snapshots of live pages.
Click on "Deploy Server".
Wait a few minutes for the server to deploy. Once ready, it will show a "Started" state.
In the chat, type @ followed by the MCP server name and your instructions, e.g., "@pagouse-mcp list the accessible elements on my current tab"
That's it! The server will respond to your query, and you can continue using it as needed.
Here is a step-by-step guide with screenshots.
By ref, not by pixel. pagouse is a page agent for coding agents: a Chromium page agent that owns an isolated browser session, reads the browser-computed accessibility tree, and — only with a grant — interacts with the page through WebDriver BiDi.
It observes by default. Mutation needs --allow-input. The optional MCP
server is observe-only and cannot be talked into typing.
$ pagouse --json snapshot --tab 1840 | jq -r '.snapshot.tree' | head
RootWebArea "Inbox" [ref_1]
textbox "Search" [ref_2]
button "Compose" [ref_3]browser_start starts Chromium with a temporary, owner-only profile and
persists only local session metadata.
WebDriver BiDi handles lifecycle, navigation, contexts and trusted input; CDP is restricted to the browser-computed accessibility tree.
A versioned --json envelope (schema: 2) is shared by CLI and MCP.
The layers and the non-goals: docs/architecture.md.
Related MCP server: DevTools Lens MCP
Linux
Python 3.13+ and uv
Chromium and a matching ChromeDriver on the host
The core itself has zero dependencies. magick is optional for --fit
downscaling.
git clone https://github.com/gianlucamazza/pagouse
cd pagouse
./install.shinstall.sh installs pagouse as a uv tool, writes the optional config if
missing, and links the agent skill. Then run pagouse browser_start.
Observe — never needs a grant:
pagouse --json tabs
pagouse --json snapshot --tab IDDrive — add --allow-input, click by ref, never by coordinates:
pagouse --json --allow-input click --ref ref_4 --tab ID --then snapshotFor credentials, configure allowlisted op:// references in
~/.config/pagouse/credentials.toml and use credential_fill. The resolved
value is never returned in the JSON envelope.
Every command and flag: docs/cli-reference.md. The envelope and error codes: docs/json-contract.md.
uv tool install 'pagouse[mcp]' provides pagouse-mcp: read-only observation
tools only. There is no mutate tool; drive the page through the CLI and a grant.
install.sh also links skills/pagouse/SKILL.md into any existing agent
skill root, so your agents get the playbook without setup.
Observation never asks permission; mutation refuses without an explicit grant. Scheme deny, origin policy, and secret redaction are enforced in the core. Tab titles, URLs, tree text, and shot pixels are untrusted input. Read docs/security-model.md before enabling input. Report vulnerabilities per SECURITY.md.
Page | For |
Install, start the managed browser, first commands | |
Every command and flag, and the MCP surface | |
The | |
| |
What problem this solves, the layers, the non-goals | |
The page grant, scheme deny, untrusted data |
MIT. See LICENSE.
This server cannot be deployed
Hosted browser for AI agents: screenshots, post-JS DOM, console, WCAG. No install, no API key.
Provides read access to your GKE and Kubernetes resources.
Read-only Search Console analytics, URL inspection, indexing diagnostics, and sitemaps.
Launch, inspect, control, and share isolated cloud browsers for your agents.
