OpenAI Codex

Add a screenshot tool to Codex in one block

One config.tomlblock points OpenAI Codex at Grabbit's hosted MCP server. It can screenshot any URL with no local process to launch and no headless browser to install.

Get your free test keySee the install

No local process

Codex talks to Grabbit as a remote Streamable HTTP server. There is nothing to npx, no STDIO process to keep alive, and no Chromium to install. Just a URL and a token.

~/.codex/config.toml
# ~/.codex/config.toml
[mcp_servers.grabbit]
url = "https://mcp.grabbit.live"
bearer_token_env_var = "GRABBIT_API_KEY"
Authenticate
# Provide the key the env var points at (a free sk_test_… key works):
export GRABBIT_API_KEY="sk_live_..."

# …or skip the key and authenticate over OAuth instead:
codex mcp login grabbit

Then put it to work:

terminal
codex "screenshot https://example.com and tell me whether the
        primary CTA is visible above the fold"

Same behavior on your laptop and in CI

Because the capture runs on Grabbit's infrastructure, a Codex task that screenshots a page behaves identically whether it runs interactively or in an automated pipeline. There is no headless browser to provision in the runner and no flaky local dependency to debug. Point it at a deployed URL and the image comes back the same every time.

Need the raw REST shape for an Agents SDK tool instead of MCP? It is the same endpoint. See the screenshot API reference.

Captures your agent's own browser can't get

Most agents can already drive a headless browser. It works on simple pages and quietly fails on the ones that matter: cookie and consent walls, bot detection, login or paywall gates, and heavy client-rendered pages. Grabbit runs the capture on hosted rendering infrastructure instead, so the shot comes back clean.

Local headless browser

  • Needs Chromium installed in the agent's runtime
  • Blank or blocked on bot-walled and consent-gated pages
  • You own the timeouts, retries, and scaling
  • Breaks in CI, sandboxes, and remote runners with no browser

Grabbit

  • One call, nothing to install or run
  • Dismisses cookie and consent banners automatically
  • Handles lazy-loaded, JavaScript-heavy, and full-page captures
  • Gets many sites a local headless browser is blocked on

Honest bound: no service captures literally every site. A handful of hard targets behind aggressive IP blocking still resist. But the cookie walls, consent gates, and JS-heavy pages that blank out a local headless browser are exactly what Grabbit is built to get.

Flat pricing, built for agents that loop

When an agent captures a page on every step, per-call metering turns into anxiety. Grabbit is one flat rate: $50 per year for 25,000 captures ($0.002 each), with prepaid packs that stack on top and never reset. No monthly quota to forfeit, no usage cliff. Test-environment keys render free placeholders, so you can wire the agent up before adding a card.

See full pricingCompare providers

OpenAI Codex FAQ

How do I add a screenshot tool to OpenAI Codex?
Add a single mcp_servers block to ~/.codex/config.toml pointing at https://mcp.grabbit.live, set a bearer token env var (or run codex mcp login for OAuth), and Codex gains a grab tool that screenshots any URL. No package to install.
Why a hosted MCP server instead of a local one for Codex?
Codex can run STDIO servers, but those need a local process and, for screenshots, a local headless browser. Grabbit is a remote Streamable HTTP server: one URL, no process to launch or keep alive, and the capture runs on Grabbit's infrastructure, so it behaves the same on your laptop and in CI.
Does it support OAuth or just an API key?
Both. Point bearer_token_env_var at an environment variable holding your sk_ key, or run codex mcp login grabbit to authenticate over OAuth. A free test key returns placeholder renders so you can wire it up first.
Can Codex screenshot any website?
It captures public URLs and handles cookie and consent banners, lazy-loaded and JavaScript-heavy pages, and many sites that block a local headless browser. Private and internal addresses (including localhost) are blocked by SSRF protection.
What tools does it add?
grab (screenshot a URL), get_grab and list_grabs (fetch or browse captures), get_usage (credits and plan), and get_pricing (a dated comparison against ScreenshotOne, Urlbox, Browserless, and others).

Give OpenAI Codex eyes in two minutes

Free test key, one install, and your agent can screenshot the web.

Get your free test keyAll agent integrations