Skip to content

AI agents

The HTTP Archive agent plugin lets AI coding agents such as Claude Code and Codex answer questions about how the web is built. It gives an agent two ways to reach HTTP Archive data:

  • Tech Report tools for technology adoption, Core Web Vitals, Lighthouse and page weight trends. These come from the Tech Report through a remote MCP server and need no Google Cloud account.
  • BigQuery skills for anything else. They teach the agent to inspect live table schemas, read these docs, preview rows for free, and write partition- and cluster-pruned SQL against the public dataset.
Terminal window
claude plugin marketplace add httparchive/httparchive-plugin
claude plugin install httparchive@httparchive

Or, inside a session, run /plugin marketplace add httparchive/httparchive-plugin, then /plugin install httparchive@httparchive.

The plugin connects the Tech Report MCP server automatically. To use the server on its own, without the skills, add it to any MCP client that supports Streamable HTTP:

Terminal window
claude mcp add --transport http httparchive https://cdn.httparchive.org/mcp
Skill Use
httparchive-docs-and-metrics Looks up these docs and discovers which custom_metrics keys exist in crawl.pages
httparchive-query-optimizer Inspects schemas, previews rows with bq head, dry-runs queries, and applies the rules in Minimizing query costs
Tool Returns
search_technologies, list_categories, list_versions Technologies, categories and versions tracked by the Tech Report
list_ranks, list_geos Rank segments and geographies available as filters
get_adoption_metrics Share of sites using a technology over time
get_cwv_metrics, get_cwv_distribution, get_geo_breakdown Core Web Vitals pass rates, histograms and per-country breakdowns
get_lighthouse_metrics, get_audits_metrics Lighthouse category scores and audit results
get_page_weight_metrics Page, JavaScript and CSS weight

The Tech Report tools work out of the box.

The BigQuery skills run the bq command-line tool, so install the Google Cloud CLI and authenticate against a project with BigQuery enabled, as described in Getting started. Queries are billed to that project. The skills read these docs with the agent’s built-in web fetch, so they need no other tools.

Every docs page is also available as Markdown at its URL plus .md, for example /docs/guides/minimizing-costs.md, and the full docs are at /llms.txt. Point any agent at these to give it HTTP Archive context.

  • “How has React adoption changed over the last two years, and how do React sites compare on Core Web Vitals?”
  • “Which countries have the best LCP pass rate for WordPress sites?”
  • “Write a query that counts how many of the top 10,000 mobile pages load a third-party font, and check its cost first.”
  • “Which custom_metrics keys describe structured data on a page?”

The plugin is open source on GitHub. Report issues or suggest new skills there.