Zenrows for ChatGPT and Codex

Live web data for ChatGPT and Codex. This plugin connects them to the Zenrows MCP server, so the model can read the current content of any public page, pull structured fields from it, run batch jobs over many URLs, crawl from a start page, and drive a hosted browser for pages that need clicks, forms or pagination. It also adds skills that tell the model which Zenrows tool fits each task and how to keep requests efficient.

What you get

• The Zenrows MCP server (https://mcp.zenrows.com/mcp). You sign in to your Zenrows account with OAuth; there is no API key to copy.
• Nine skills:

Skill	Use it for
using-zenrows	Overview: tools, defaults, errors and plan limits
getting-started	First-run checks and troubleshooting (the plugin's onboarding skill)
scrape-webpage	Reading one page
extract-structured-data	Named fields or JSON from one page
batch	A list of five or more URLs as one job
crawl	Many pages behind one start URL, with the Crawl tools
map	Listing a site's URLs from its sitemap without fetching them
browser-automation	Clicks, forms, pagination and multi-step flows
capture-api-calls	Finding the API calls a page makes

Requirements

A Zenrows account. Requests use credits on your Zenrows plan. You can sign up here.

Privacy

The plugin sends the URLs and options you ask the model to fetch to Zenrows, which processes them under the Zenrows privacy policy.

Support

• Documentation: docs.zenrows.com
• Contact: success@zenrows.com

For maintainers

The package follows OpenAI's plugin package format: a root plugin.json (Agent Plugins schema, with the listing, review and publication fields under extensions.com.openai), a root mcp.json, skills/ and assets/.

node scripts/validate.mjs    # checks the documented submission limits
bash scripts/package.sh      # validates, then builds dist/zenrows-<version>.zip

scripts/validate.mjs encodes the limits from OpenAI's manifest field reference and submission error reference. CI runs it on every push.

Things to know before submitting:

• Reviewer credentials never go in this repo. OpenAI rejects test_credentials and reviewer_instructions in the package; they are entered in the portal's Review details form.
• Run the review test cases with the reviewer test account before submitting. OpenAI asks for it. Update expected_behavior to match what actually happens.
• demo_recording_url must open without signing in. It is required for MCP review.
• mcp.json declares OAuth under extensions.com.openai.auth, as OpenAI's authentication reference says. The generic Agent Plugins MCP schema doesn't list extensions on a server; OpenAI's own examples combine both, so this follows OpenAI.
• The tool scan reads the live server. Every tool on mcp.zenrows.com needs readOnlyHint, destructiveHint and openWorldHint before submission (ZenRows/zenrows-mcp#43).
• Bump version and add a CHANGELOG.md entry for every upload. OpenAI requires a new version for each release.

License

MIT