[tedi-z]

What this PR does

• Plugin name: zenrows
• Type: remote source
• Source URL + pinned SHA: https://github.com/ZenRows/grok-zenrows.git @ ac8e033a2a87f0664f57009c25d98742cf209aab (tag v0.1.2)
• Homepage: https://www.zenrows.com

Zenrows gives Grok live web data: fetch a page as markdown, 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, logins or pagination.

Ownership

• I own this plugin or have the right to distribute it.
• The source repo is published under our official org.

Checklist

• Added exactly one entry to .grok-plugin/marketplace.json (valid JSON, kebab-case name).
• Remote source pins a full 40-char lowercase commit sha, public and reachable.
• Regenerated .grok-plugin/plugin-index.json.
• python3 scripts/validate-catalog.py passes locally.
• python3 scripts/generate-plugin-index.py --check passes locally.
• homepage + clear description set.
• License is stated (MIT, in the manifest and LICENSE).

Security

• No curl | bash, remote-code download/exec, or postinstall RCE.
• No reading/exfiltration of secrets, tokens, .env, or env vars.
• Hooks and MCP scope are least-privilege.

Network endpoints this plugin calls (and why):

• mcp.zenrows.com — the hosted MCP server itself, over HTTP. Every tool call goes here.
• app.zenrows.com — OAuth only: authorization, token exchange and dynamic client registration.

Pages the user asks for are fetched by Zenrows, so the plugin makes no direct request to the target site.

Credentials/permissions it requires (and why):

An OAuth token for the user's Zenrows account, scope api, obtained through the browser flow on first connection and stored by Grok in ~/.grok/mcp_credentials.json. The plugin never reads or writes that file.

No API key, no environment variables, no hooks, and no code Grok runs. What Grok loads is a .mcp.json pointing at the hosted server plus nine Markdown skills. The repository also carries scripts/validate-plugin.py, a development-time check run by our own CI; Grok neither loads nor executes it.

Two skills suggest saving large result sets under ./.zenrows/ in the working directory, and only when the user asked for the data. Grok's own file-write permissions apply.

Notes for reviewers

Submitted by the Zenrows team (contact: success@zenrows.com).

We ran both gates locally against this SHA: validate-catalog.py and generate-plugin-index.py --check both pass, and the diff is additive only (63 insertions, 0 deletions).

We also installed and exercised the plugin end to end in Grok Build 1.0.70:

• grok plugin install ZenRows/grok-zenrows@v0.1.2 --trust
• /skills lists the nine zenrows-* skills; /mcps shows the server ready
• OAuth completes in the browser and the token is stored
• Read a page: scrape returned the four out-of-stock products on a test catalog
• Structured fields: extract returned the 16 products and prices on page 1, without paging further
• Browser: browser_navigate → browser_get_accessibility_tree → browser_batch → browser_close found a price behind a "Load more" click
• Negative control: an unrelated prompt triggered no Zenrows call

Skill names carry a zenrows- prefix because Grok loads skills from every enabled plugin into one namespace; bare names like batch or map would collide with plugins already in this catalog.