Codex CLI fetch URL: how to make Codex read a web page

So a link in your prompt is often not enough. Codex can search for the page, or it can try curl. In the default cached mode, the search gives it what the index holds, which can be old or can be only a part of the page. curl gives it raw HTML when the site answers, and an error page when the site does not.

Anyfetch is a free, open-source MCP server from the AliasMode team. It gives Codex three tools: web_fetch, web_fetch_many, and web_search. Pick one of the three ways.

1. 1

   With Node.js: one command, no install

   npx downloads the program for your system at the first start.

   Step 1

   codex mcp add anyfetch -- npx -y @aliasmode/anyfetch mcp

2. 2

   With the installer

   The installer puts one program in ~/.local/bin and runs codex mcp add for you. It also connects Claude Code, Gemini CLI, Cursor, and the other agents it finds.

3. 3

   By hand in config.toml

   Add the server to ~/.codex/config.toml. Use this when you manage the file yourself.

   Step 3

   [mcp_servers.anyfetch]
   command = "anyfetch"
   args = ["mcp"]

curl -fsSL https://aliasmode.github.io/anyfetch/install.sh | sh

irm https://aliasmode.github.io/anyfetch/install.ps1 | iex

1. Run codex mcp list. The list must show anyfetch.
2. Start a new Codex session. Tools are loaded at the start of a session.
3. Ask for a page that curl cannot read: “Use web_fetch to read https://www.reuters.com/technology/ and list the first three headlines.”
4. Codex calls web_fetch and answers from the page. If it runs curl in place of the tool, name the tool in your prompt or in AGENTS.md.

Do you want Codex to use only Anyfetch for the web? Turn the hosted search off in ~/.codex/config.toml. Anyfetch has its own web_search, which uses Brave, Yahoo, and DuckDuckGo with no API key.

web_search = "disabled"

• Markdown in place of HTML. One Wikipedia article is 236,000 bytes of HTML from curl and 46,000 characters of Markdown from Anyfetch.
• Only the parts that answer a question, when Codex gives a query with the URL. The default limit is then 5,000 characters.
• PDFs as text, page by page.
• A Reddit thread with every comment, a Stack Overflow question with its answers, and full Reuters and Bloomberg articles.
• An error code when a page stays closed, such as challenge_required or http_error, so Codex does not guess.

anyfetch search "tokio select" --site docs.rs --time year
anyfetch fetch https://docs.python.org/3.15/whatsnew/3.15.html --query "sampling profiler rate"

Do you not want an MCP server? Put the two commands in AGENTS.md, and Codex runs them in its shell. The shell needs network access for that.

## Web access
- Search the web: `anyfetch search "<query>"`
- Read pages: `anyfetch fetch <url> [<url>...] --format markdown`

Anyfetch reads public pages. It does not log in, click, or run page scripts. For a task with an account, a form, or a web app, Codex needs a real browser. AliasMode gives Codex browser profiles with saved sessions through its own MCP server. Use Anyfetch to read, and AliasMode to act.