Does Claude Code Support AGENTS.md?

No. Claude Code reads CLAUDE.md natively. Here is the official symlink and import workaround, and which coding agents in our index already read AGENTS.md on their own.

By The Agents Index Editorial, Research desk6 min readPublished 2026-08-19Re-verified 2026-08-19

CoversClaude Code · AGENTS.md · CLAUDE.md · Coding agents · Compatibility

No. Claude Code does not read AGENTS.md natively. It reads a file called CLAUDE.md instead, and Anthropic's own documentation says so in exactly those words. If your repository already carries an AGENTS.md for other tools, Claude Code will not pick it up on its own. Two officially supported fixes take under a minute, and a third moved a whole other tool's setup over in one command as of a recent release. All three are below, along with which coding agents in our own index already read AGENTS.md without any of this.

What Claude Code actually reads

Claude Code loads project instructions from CLAUDE.md (or .claude/CLAUDE.md), plus a personal ~/.claude/CLAUDE.md and, for teams on managed policy, an organization-wide file. None of those paths is AGENTS.md. Anthropic's memory documentation is direct about it: "Claude Code reads CLAUDE.md, not AGENTS.md."

This puts Claude Code in a specific minority. AGENTS.md is an open format for agent instructions, stewarded by the Agentic AI Foundation under the Linux Foundation, and it was built with input from OpenAI Codex, Google's Jules, Cursor and Factory among others. The format's own registry lists more than 30 supporting tools. Claude Code reading a Claude-specific file instead is a deliberate holdout, not an oversight: the same documentation page walks through how to bridge the two.

Two official ways to point Claude Code at AGENTS.md

Both come straight from Anthropic's docs, not from a community workaround, so neither breaks on the next release.

Symlink it. If you don't need Claude-specific instructions on top of what's already in AGENTS.md, run this once from your project root:

ln -s AGENTS.md CLAUDE.md

The command prints nothing on success. Confirm it worked by running /context in your next Claude Code session and checking that CLAUDE.md appears under Memory files. On Windows, symlinks need Administrator privileges or Developer Mode, so use the import method below instead.

Import it. If you want Claude-specific rules alongside the shared file, create a CLAUDE.md that imports AGENTS.md and adds to it:

@AGENTS.md

## Claude Code
Use plan mode for changes under src/billing/.

Claude Code loads the imported file at session start, then appends whatever you write below it, so the two files stay in sync without duplicating a word.

A third way: /import, for a one-time copy

Claude Code v2.1.213 added an /import command that reads a supported coding agent's configuration, including its AGENTS.md, and copies the relevant pieces into CLAUDE.md, along with MCP servers, commands, subagents and skills where they carry over. It's a one-time migration rather than a live link: run it once when you're switching a project onto Claude Code, not as an ongoing sync. Running /init without that command still reads AGENTS.md for context if you set CLAUDE_CODE_NEW_INIT=1, alongside Cursor rules, Copilot instructions, Devin, Windsurf and Cline configuration files.

Which coding agents in our index already read AGENTS.md

Of the coding agents we track, these are named on the AGENTS.md project's own list of supporting tools. No symlink or import is required, because they read the file directly.

Agent	Reads AGENTS.md natively?
Cursor	Yes, also usable as a drop-in alternative to `.cursor/rules`, including nested files per subdirectory
GitHub Copilot	Yes, for its coding-agent instructions specifically. The nearest `AGENTS.md` in the directory tree takes precedence
OpenAI Codex CLI	Yes, Codex is one of the tools that originated the format
Aider	Yes
OpenCode	Yes
Zed	Yes
Warp	Yes
Devin	Yes
Junie	Yes
Windsurf	Yes
Augment Code	Yes
Factory	Yes
Goose	Yes
Claude Code	No, reads `CLAUDE.md`; use the symlink or import above

The pattern in that table is the whole story: a single AGENTS.md at the root of a repository already reaches most agents a developer is likely to run side by side. Claude Code is the one that needs a one-line bridge, and Anthropic ships that bridge itself rather than leaving it to a community script.

Which fix to use

Symlink if AGENTS.md already says everything you want Claude Code to know: it's one command and there's nothing to keep in sync by hand. Import if you want anything Claude-specific, like a plan-mode rule scoped to one directory, layered on top without touching the shared file other agents read. Reach for /import only when you're moving a project onto Claude Code for the first time and want its existing AGENTS.md, commands and MCP servers carried over in one pass, not as a recurring sync step.

Screenshot of Claude Code — 13 agents this guide names, as their own sites render them — captured by theagentsindex.com. Each tile opens the full listing.

Screenshot of Cursor — 13 agents this guide names, as their own sites render them — captured by theagentsindex.com. Each tile opens the full listing.