Skip to main content
Glama
local-chrome-cdp-probe-mcp

local-chrome-cdp-probe-mcp

local-chrome-cdp-probe-mcp

Real-browser measurement + screenshots through your local Chrome via the Chrome DevTools Protocol. Zero npm dependencies (Node >= 20 global fetch/WebSocket), zero Python deps beyond the MCP SDK. For when Playwright is not cached and agent-browser is not installed, but the bug report is "this button looks squashed on my screen".

Why this exists — the traps it removes

Trap

Symptom

Handled by

/json/new with GET

plain text "Using unsafe HTTP verb GET..." then .json() explodes

probe uses PUT (Chrome >= 111 contract)

Host sandbox kills Chrome's sandbox

renderer dies, DevTools ws closes with 1006 immediately

--no-sandbox --disable-dev-shm-usage baked in

Node WebSocket sends an Origin header

DevTools rejects the upgrade, 1006

--remote-allow-origins=* baked in

Retina display doubles physical pixels

measured sizes 2x of CSS px

--force-device-scale-factor=1 baked in

httpOnly login cookies unreachable from scripts

"just set document.cookie" fails

login is a fetch POST executed INSIDE the page - the browser's cookie jar applies

Modal overlays ruin screenshots

shot full of forced-password-change mask

your measure_js can remove the mask before capture

Related MCP server: cliqueja

Install & run

pip install -e .            # mcp only
node --version              # must be >= 20
python server.py            # stdio transport
{
  "mcpServers": {
    "local-chrome-cdp-probe": {
      "command": "python",
      "args": ["/path/to/local-chrome-cdp-probe-mcp/server.py"]
    }
  }
}

Optional env: CHROME_PATH (any Chromium), NODE_BIN (default node), PROBE_TIMEOUT (wrapper timeout, default 60s; per-probe --timeout-ms default 30s).

Smoke test

python test_smoke.py    # real Chrome round trip on a local file:// page

License: MIT.

Available Tools

2 tools
probe_availableA

Preflight: check node + the bundled probe script + Chrome detection.

Inputs: no named parameters.

probe_pageA

Measure and/or screenshot a page in real local Chrome.

Args: url: full URL to probe (http/https/file) measure_js: in-page JS expression (IIFE returning an object) evaluated after load; its value comes back as measurements. Example that measures one element's real text width: (() => { const el = document.querySelector('#target'); if (!el) return null; const b = el.getBoundingClientRect(); const cs = getComputedStyle(el); return { x: Math.round(b.x), y: Math.round(b.y), w: Math.round(b.width), h: Math.round(b.height), usable_text_width: Math.round(el.clientWidth - parseFloat(cs.paddingLeft) - parseFloat(cs.paddingRight) - 2) }; })() viewport_w / viewport_h: emulated viewport (also set via Emulation.setDeviceMetricsOverride, so it holds regardless of the Chrome window size) screenshot: capture a PNG and return it as base64 login: "user:pass" - a JSON POST login executed IN the page context before measuring (httpOnly cookies are fine - the page fetch carries them). Requires the target app to have a JSON login API. login_url: login endpoint path, default /api/auth/login login_body: JSON template with %u / %p placeholders, default {"username":"%u","password":"%p"} timeout_ms: overall probe timeout

Returns: {ok, origin, viewport, login_status, measurements, png_base64}

Inputs: url (required), login, login_url, login_body, +5 more.

Tool Schema Changelog

Recent tool additions, removals, and schema changes observed during successful MCP inspections.

Loading tool schema changelog…

    TDQS

    A3.9/5.0

    Scored across 2 tools

    Disambiguation5/5

    The two tools have clearly distinct roles: probe_available is a preflight/environment check (node, bundled script, Chrome detection) while probe_page performs the actual measurement/screenshot work. There is no overlap in purpose, so an agent can pick correctly without hesitation.

    Naming Consistency5/5

    Both tools use the same snake_case, 'probe_' prefixed pattern (probe_available, probe_page), which reads as a deliberate, predictable namespace. No mixing of conventions or verb styles.

    Tool Count3/5

    Two tools is on the thin side, though the server's scope (a local Chrome CDP probe) is intentionally narrow. Much of the capability is folded into probe_page's many parameters, so the count is defensible but borderline minimal.

    Completeness4/5

    The surface covers preflight, navigation, viewport emulation, authenticated JSON login, arbitrary in-page measurement, and screenshots, which is solid for the stated purpose. Gaps exist around multi-page/batch probing, wait conditions (network idle, selector waits), and console/error capture, but the free-form measure_js escape hatch lets agents work around most of them.

    Maintenance

    ActivitySlowing
    ResponsivenessNo recent community issues

    Related MCP Connectors

    Related MCP Servers

    • A
      license
      Not graded
      quality
      D
      maintenance
      Enables MCP clients to drive a real, logged-in Chrome browser for web automation tasks like navigation, clicking, typing, and screenshotting.
      2 npm
      MIT
    • A
      license
      Not graded
      quality
      D
      maintenance
      Enables fast browser automation on a real Chrome tab via a local hub, providing MCP tools for clicking, filling, snapshotting, and navigating without launching a full browser instance.
      MIT
    • A
      license
      B
      quality
      C
      maintenance
      Exposes a live Chrome browser to any MCP client through a single local endpoint, enabling real-tab control, DOM interaction, cookies and session use, network/console capture, PDF export, and CDP-powered debugging.
      28
      MIT
    • A
      license
      Not graded
      quality
      B
      maintenance
      Enables AI agents to see and control the user's real Chrome/Brave/Edge profile over MCP, so they can read pages, click, type, take screenshots, audit layouts, debug CSS, and scrape paginated or infinite-scroll data. Because it drives the normal browser via the DevTools protocol with real input events, it works on modern JavaScript apps and on sites where the user is logged in.
      17 npm
      1
      MIT