Codex is OpenAI's coding agent, powered by ChatGPT. It supports MCP through a Custom MCP entry in its settings. Once connected, Codex can call the same ZoomInfo tools that ship with Claude Desktop and ChatGPT: search, enrichment, similar companies, intent, research.

Setup is short. There is no DevPortal app, no OAuth callback to host, and no CLI command. You paste the ZoomInfo MCP URL into Codex's Custom MCP form and complete the ZoomInfo login.

What You Need Before You Start

A Codex account with Custom MCP support. OpenAI exposes MCP as a Custom MCP form in Codex Settings. If you don't see it, your plan or workspace may not have it enabled yet.

A ZoomInfo subscription with bulk data credits. Any ZoomInfo package qualifies, but your account must have bulk data credits enabled. Recurring monthly credits are not supported by MCP. Check with your ZoomInfo admin if you're unsure. See the full requirements.

Step-by-Step Setup

Open Codex Settings

From inside Codex, go to Settings, then Custom MCP.

Add a new MCP server

Click Add (or the equivalent new-entry button) and fill in:

Name: ZoomInfo
Transport: Streamable HTTP
URL: https://mcp.zoominfo.com/mcp

Save and authenticate

Click Save. Codex prompts you to sign in with ZoomInfo. Use your normal ZoomInfo credentials or SSO.

Verify the connection

Start a new Codex session and ask: "What ZoomInfo tools do you have access to?" Codex lists the tools it sees from the MCP server.

Verifying the Connection Works

Try a question that needs ZoomInfo data:

How many employees does Snowflake have?

or:

Look up the CEO of Datadog.

If the connector is healthy, Codex calls the matching ZoomInfo tool and returns live data. The tool invocation appears inline.

For a chained smoke test:

Find 5 cybersecurity companies in New York with 200-500 employees, then find the VP of Engineering at each.

That exercises Lookup, Search Companies, and Search Contacts in sequence.

Working With Tools in Codex

Codex selects tools the same way other MCP clients do: it reads the tool descriptions and picks the right one for your request. You don't need to name the tool.

A few things worth knowing for the coding-agent context:

Tool results land in your session context. Direct tools (Search, Enrich, Lookup) return data into Codex's context window. Large result sets eat context fast, so narrow filters before running a search across thousands of records.

Research tools run sub-agents. Account Research and Contact Research synthesize large payloads before returning a summary. They cost more credits per call but keep your context manageable.

Bulk credits apply the same way. Each enrichment record consumed counts against your bulk credits regardless of which client made the call.

Troubleshooting

Codex doesn't show a Custom MCP option. Your Codex plan or workspace may not have Custom MCP enabled. Check with your OpenAI admin or workspace owner.

Authentication fails on Save. Confirm you're signing in with your personal ZoomInfo credentials, not a shared or admin-only account. If your ZoomInfo session has expired, the connection won't complete.

Codex doesn't see any ZoomInfo tools. Open Settings → Custom MCP and confirm the entry is listed and the URL is exactly https://mcp.zoominfo.com/mcp. Remove and re-add the entry if the auth state looks stale.

Credit errors on enrichment or research calls. MCP requires bulk data credits. Confirm with your ZoomInfo admin that your account has bulk credits enabled, not recurring monthly credits.

What's Next

With ZoomInfo wired into Codex, you can pull account lists, enrich contacts, and prep deals in the same session you're writing code. For the full tool reference, see the MCP Tools documentation. For the equivalent setup in Anthropic's CLI, see How to Connect ZoomInfo to Claude Code Using MCP.