MCP Web Search & Scrape in Codex CLI

Connect OpenAI Codex CLI to Firecrawl through the hosted OAuth MCP server.

Quick setup#

1. Add Firecrawl#

codex mcp add firecrawl --url https://mcp.firecrawl.dev/v2/mcp-oauthcodex mcp login firecrawl

The URL is a client configuration value, not a page to open directly in a browser. Codex opens the Firecrawl sign-in flow, where you choose a team and approve the connection.

2. Start Codex and verify#

codex

Enter /mcp and confirm that firecrawl is connected. If you changed an existing Firecrawl connection, start a new Codex session before retrying.

Try:

Search the web for the latest Next.js App Router release notes and summarize the sources.

Scrape https://docs.firecrawl.dev and list the top-level sections.

Other connection modes#

• Try keyless MCP for Search, Scrape, and Parse without an account or key.
• Configure an API key with bearer_token_env_var for unattended use. Keep the secret outside agent chat and source control.
• Run MCP locally only when you need a local process or a self-hosted Firecrawl API. The local package requires Node.js 22 or newer.

Optional: literature research#

For paper and literature work, the research-index skill routes those queries to the firecrawl_research_* tools and the Research Index instead of general web search:

npx skills add firecrawl/skills@firecrawl-research-index

Troubleshooting#

• Codex does not see the tools: run codex mcp list, confirm that only one firecrawl entry exists, and restart Codex.
• OAuth returns 401: run codex mcp login firecrawl and complete sign-in. A bare request to the OAuth server is expected to return 401 before authentication.
• An API-key connection returns 401: replace the key in the environment that starts Codex, then start a new session.