Firecrawl Codex plugin

OpenAI Codex plugin that bundles Firecrawl CLI workflow skills (search, scrape, crawl, map, interact, agent, parse, download) and the Firecrawl MCP server for tool-style access.

Canonical skill sources live in firecrawl/cli (cli/skills/). This repo vendors copies for the Codex plugin distribution; refresh when CLI skills change materially.

Requirements

• Node.js 18+ (for npx firecrawl-mcp when using bundled MCP).

• Firecrawl CLI installed and authenticated for skill workflows that shell out to firecrawl:

  npm install -g firecrawl-cli
  firecrawl login --browser

  Or set FIRECRAWL_API_KEY in your environment.

Install in Codex (local marketplace)

Until the plugin is listed in the official Codex Plugin Directory, wire a repo or personal marketplace as described in the OpenAI docs.

Option A: Git clone next to your project

1. Clone this repository (for example into plugins/firecrawl-codex-plugin at the root of the repo where you use Codex).

2. Add .agents/plugins/marketplace.json at that repo root (see examples/marketplace.json). Point source.path at the plugin folder with a ./-prefixed path relative to the marketplace file’s directory.

3. Set MCP credentials in ~/.codex/config.toml so the bundled server can authenticate (do not commit secrets):

   [mcp_servers.firecrawl.env]
   FIRECRAWL_API_KEY = "fc-YOUR-API-KEY"

   Alternatively use a remote MCP URL and disable or omit the bundled MCP entry in your Codex config if you prefer.

4. Restart Codex. Open Plugins (Codex app) or run /plugins in Codex CLI, pick your marketplace, install Firecrawl, enable the plugin.

Option B: Codex CLI marketplace add

From the OpenAI docs:

codex plugin marketplace add ./path-to-marketplace-root

Use a marketplace JSON that references this plugin directory (same shape as examples/marketplace.json).

Bundled MCP

.mcp.json runs npx -y firecrawl-mcp. You must provide FIRECRAWL_API_KEY via Codex MCP server environment configuration.

Smoke checks (after install)

In a new Codex thread, try:

1. “Search the web for the latest Firecrawl changelog and summarize.”
2. “Scrape https://docs.firecrawl.dev and list top-level sections.”
3. “Use Firecrawl MCP to call firecrawl_scrape on https://example.com.” (if MCP is enabled)

CLI-only checks from a terminal:

firecrawl --status
firecrawl search "firecrawl web scraping" --limit 3 --json

The _local/ directory is gitignored; use it in your clone for any markdown or notes you do not want committed.

License

MIT — see LICENSE.