How to Add Image Search to Codex

Codex lives in the terminal, but the work it helps with sometimes has a visual dimension — UI components, asset selection, design references, brand research. Adding Google Image Search via AgentPatch lets Codex pull visual information during a session without you having to gather it manually.

Why This Matters

Development tasks involving front-end work, documentation, or brand alignment often require visual references. Finding those references typically means opening a browser, searching, deciding what’s relevant, and bringing the information back to your coding session.

When Codex has image search as a tool, it can handle that step itself. You describe what you’re looking for and it finds it — the same way it would search for a code example or documentation page.

Google Image Search also supports filters (image type, size, color, usage rights), which the tool exposes. That’s useful when you specifically need something like a transparent-background logo or a photo within a particular usage license.

Setup

Add AgentPatch to ~/.codex/config.toml:

[mcp_servers.agentpatch]
url = "https://agentpatch.ai/mcp"
bearer_token_env_var = "AGENTPATCH_API_KEY"

Then set your API key:

export AGENTPATCH_API_KEY=your_api_key

Replace your_api_key with your actual key from the AgentPatch dashboard. Codex discovers all AgentPatch tools automatically on next start.

Example

You’re building a component library and want to document what each component should look like in practice:

“I’m documenting a data table component. Can you find a few screenshots of well-designed data tables with filters and pagination as visual reference?”

Codex searches Google Images:

Searching Google Images for "data table UI component filters pagination design example"...

Found relevant results:

1. Airtable grid view — inline filters, sticky header, pagination controls
   [image URL]

2. Linear issues list — clean row design, status badges, dense but readable
   [image URL]

3. Notion database view — toggleable column types, sort indicators
   [image URL]

Common patterns across these: sticky column headers, row hover states, pagination
or infinite scroll (not both), filter chips above the table rather than inline.

Want me to use these as reference while building out the component spec?

Codex found relevant examples, extracted the common design patterns, and offered to apply them directly to the task. That’s the kind of visual research loop that normally costs several minutes of manual work.

For more practical tasks, image search is also useful for finding public domain photos for test content, checking what competitor product pages look like, or locating official logos and brand assets.

Wrapping Up

Image search rounds out Codex’s research capabilities beyond text. One AgentPatch connection gives access to Google Image Search, Google Search, Bing Search, and everything else on the marketplace — no extra setup per tool. Explore the full list at agentpatch.ai.

← Back to blog