Best Web Data Skills and Plugins for ChatGPT in 2026

On October 30, 2026 Firecrawl published a guide to ChatGPT skills that ends with five example skills built on its plugin. It is a useful prompt to ask a narrower question: if you want a skill in ChatGPT that works from live web data, what can you install, and what do the other web-data vendors offer? This post answers that for the four we compare elsewhere.

Skill, plugin, app: three different things

• Skill. A folder with SKILL.md (and optionally scripts or reference files). ChatGPT reads the name and description first and loads the full instructions when your request matches, or when you call it with an @ mention. You can create one from chat, write it in the editor, or upload a folder. OpenAI scans uploads, and some may be flagged for review or blocked.
• Plugin. An installable bundle from the Plugin Directory. Firecrawl's is one; Codex shares the same directory.
• App (custom connector). An MCP server you add yourself, which in ChatGPT means turning on Developer Mode and creating an app with the server's URL.

The skill is the cheapest part to write and the least useful alone. A "competitor price check" skill with no way to read a price page is a checklist. Pair it with something that returns data.

Who can use skills

OpenAI's help article says personal skills are generally available to ChatGPT Business, Enterprise, Healthcare and Edu users, with admins controlling who can create, use, upload, share and install them through Permissions & roles. Sharing happens from the Skills tab of the Plugin Directory, by person, group or link. The same pages say skills were off by default for Enterprise and Edu, with OpenAI planning to turn them on by default from July 23, 2026 for Enterprise workspaces that had not opted out.

None of the OpenAI pages we could read mention Free, Go, Plus or Pro. Third-party guides disagree: one says the personal Plus and Pro plans can upload skills, another says they cannot. OpenAI's pages are the authority, and they change, so treat the sidebar as the test. Firecrawl's own post says skills became generally available on July 9, 2026; we did not find that date on an OpenAI page, so we do not rely on it. We could not open the help article directly (it returned 403 to our fetcher) and read it through search excerpts, so check it before quoting a plan.

The contenders

Firecrawl: a plugin with its own skills

Firecrawl announced an official ChatGPT plugin on August 5, 2026: open its page in ChatGPT, click Install, and connect your account. The announcement lists search, scrape, crawl, interact, monitor and Alexandria (discovery across its data-provider index) and says the plugin ships with ready-made skills for common tasks; it does not say which plans are required or name those skills. The October 30 guide then shows five skills you write yourself that call the plugin: fantasy-lineup-check, account-brief, filing-brief, podcast-sponsor-research and hiring-signals. This is the most complete answer of the four, and it is Firecrawl's own account of itself: we have not installed the plugin. Choose it when you want a plugin that is already in the directory and a set of starter skills.

Bright Data: a custom app, with the token in the URL

Bright Data documents ChatGPT in its MCP docs. The steps: use the plus button to add a source, choose Connect more, open Advanced settings and enable Developer Mode, then create an app named whatever you like, with the server URL https://mcp.brightdata.com/mcp?token=YOUR_API_KEY and authentication set to none. Then tag the app in a chat. Two things follow from that design. The page names no plan requirement, so what Developer Mode needs on your plan is an open question. And the API key sits in the URL you paste, so it is visible to anyone who can see the app's settings. It offers no skill or plugin of its own for ChatGPT. Choose it when the target is a protected site and you accept a shared secret in a URL.

Zyte: no ChatGPT path in its docs

Zyte launched a hosted MCP server at https://mcp.zyte.com/v1/mcp on September 30, 2026, with OAuth sign-in. Its docs give install steps for Claude Code, Codex CLI, GitHub Copilot CLI, VS Code and claude.ai, and a generic JSON block for other clients; ChatGPT is not named. The generic remote server could be added as a custom app, but we did not try it and Zyte does not claim it. Its fetch_http, fetch_page, extract_from_http and extract_from_browser tools would be what a skill calls. Choose it if you already pay for Zyte API or Scrapy Cloud and are willing to test the connection yourself.

Crawlora: skills you upload, an app you add

Crawlora's open-source crawlora-skills repository holds one folder per research task (product prices, SERP keywords, jobs, finance, app reviews and more), each with a SKILL.md. They follow the same Agent Skills format, so a folder can be zipped and uploaded through Skills, Create, Upload from your computer. Those skills bundle a shell script that calls Crawlora's REST API with your key. Whether ChatGPT lets an uploaded skill run a script and reach the network is something we did not find documented, so we have not assumed it. The safer pattern is the one below: add Crawlora's hosted MCP server as an app, and write a skill that tells ChatGPT which of its tools to call.

For ChatGPT, prefer a narrow view of the server over the full list of 4,204 tools: https://mcp.crawlora.net/mcp/family/web serves 9 web search, scrape, crawl and map tools, and https://mcp.crawlora.net/mcp/agent serves 3 tools that search the whole catalog. The server advertises OAuth 2.1 with Dynamic Client Registration. One thing we saw on November 2: the protected-resource metadata is served at https://mcp.crawlora.net/mcp/.well-known/oauth-protected-resource, while the root /.well-known/oauth-protected-resource and the path-inserted /.well-known/oauth-protected-resource/mcp forms both return 404. A client that follows the WWW-Authenticate hint will find it; one that probes only the standard locations may not, and we have not tested ChatGPT. Choose it when the skill's job is "get this record from a named platform", and test the sign-in before you share the skill.

A skill that uses live data

This is our template for a data-backed skill, not a file we have run in ChatGPT. Adjust the tool names to whichever app you installed:

---
name: competitor-price-check
description: Use when the user asks how a product's price compares across retailers. Pulls live prices from the connected web-data app and returns a table.
---

# Competitor price check

1. Ask for the product name or URL if the user has not given one.
2. Use the connected web-data app (search first, then scrape the top retailer pages). Do not answer from memory.
3. Return a table: retailer, price, currency, date checked, link.
4. If a page could not be read, list it as "not retrieved" instead of guessing.
5. State the date and the source apps used at the end.

Skills trigger on the description, so make it say when to use the skill, and name the app category rather than a specific tool if you want it to survive a vendor swap. Custom GPTs are the thing skills replace for many teams; Firecrawl's post says they retire on December 11, 2026, which we have not verified with OpenAI.

How to choose, by job

• Want a plugin already in the directory, plus starter skills? Firecrawl.
• Need a protected site, and fine with a token in a URL? Bright Data's custom app.
• On Zyte already? Test the generic remote server before building a skill on it.
• Want records from many named platforms behind one skill? Crawlora, through a narrow endpoint, after you have confirmed the sign-in in your workspace.
• Not on a plan that can create skills? Use the plugin or app directly in chat; you lose reusable instructions, not data access.

Give your ChatGPT skills live web data

One API key, billed only on success. 2,000 free credits a month, no card.

Set up Crawlora MCPBrowse the API catalog

Sources

Related reading

• Best Web Data Connectors for Claude in 2026 · Best Web Data Plugins for Codex and Cursor in 2026: the same comparison for the other two clients with plugin directories.
• Best AI Agent Skills Marketplaces in 2026: where shared skills live.
• Crawlora vs Firecrawl · Crawlora vs Bright Data · Crawlora vs Zyte.
• Give your AI agent live web data with MCP: how MCP works and how to connect in three minutes.