Claude-Code
The user-agent Claude Code sends when a developer asks it to fetch a URL. Its audience is developers, not consumers.
Reference reviewed by Sona
The operator publishes little or no documentation for this token. Details here come from community crawler registries and observed behavior, not vendor confirmation.
| Operator | Anthropic |
|---|---|
| Powers | Web fetches from the Claude Code CLI |
| Purpose | User-triggered fetch |
| User-agent token | Claude-Code |
| Respects robots.txt | Yes |
Claude-Code identifies web fetches made by Anthropic's Claude Code CLI - when a developer working in the tool asks it to read a documentation page, an issue, or any other URL, the request goes out under a claude-code/ user-agent carrying the CLI's release version.
One clarification on provenance: Anthropic's crawler support article covers ClaudeBot, Claude-User, and Claude-SearchBot, and does not list Claude-Code among them. The user-agent is real and observable in logs, but it is documented as a tool identifier rather than as a member of Anthropic's crawler family, so treat the crawler-style framing found in some directories with a little caution.
The audience is what makes this token worth a separate decision. Every fetch is a developer, inside their editor, actively working on something - the narrowest and most technical population of any bot here. If you publish API documentation, SDK references, or developer guides, this traffic is your target reader arriving with maximal intent. If you publish consumer content, you will barely see it.
The version segment changes with each CLI release, so match on the claude-code/ prefix or on the Claude-Code robots.txt token rather than pinning a rule to a full string. Anthropic's published crawler IP list covers its documented bots; because this is a locally-run developer tool, requests originate from the developer's own machine and network rather than from Anthropic infrastructure.
How Claude-Code behaves
- Runs locally, so requests come from the developer's own network - not from Anthropic infrastructure.
- Version segment changes with each CLI release; match the claude-code/ prefix.
- Single-page fetches of docs, issues, and references during active development work.
- Not listed in Anthropic's crawler support article, which covers only its three documented bots.
Full user-agent string
claude-code/2.0
Allow Claude-Code
Developers using Claude Code can pull your docs into their working sessions - decisive if developers are your audience.
User-agent: Claude-Code Allow: /
Block Claude-Code
You want no tool-driven programmatic access at all, even single-page fetches a developer explicitly requested.
User-agent: Claude-Code Disallow: /
How to verify Claude-Code
No published verification method
This one is structurally unverifiable, and for an interesting reason: Claude Code runs on the developer's own machine, so fetches originate from their network rather than from Anthropic's. Anthropic's published crawler IP list at claude.com/crawling/bots.json will not match, and should not - a hit from a residential or corporate address is the expected pattern here, not a red flag. There is no way to distinguish a genuine Claude Code fetch from any other client sending the same header.
Check an IP against this botCommonly confused with Claude-Code
Claude-User covers fetches from Claude's chat interface, running on Anthropic's infrastructure. Claude-Code comes from a developer's local machine.
ClaudeBot is a bulk training crawler on Anthropic infrastructure. Claude-Code is a locally-run developer tool making one-off fetches.
Claude-Code FAQs
Why doesn't Claude-Code traffic match Anthropic's published IP list?
Because Claude Code runs on the developer's own machine, so fetches originate from their network rather than Anthropic's. A request from a residential or corporate address is expected here, not a sign of spoofing.
Is Claude-Code one of Anthropic's official crawlers?
Not as documented. Anthropic's crawler support article covers ClaudeBot, Claude-User, and Claude-SearchBot only. The claude-code/ user-agent is real and observable, but it identifies a developer tool rather than a crawler in that family.
Should I allow it?
If developers are your audience, yes - every fetch is a developer actively working who wants your documentation. If you publish consumer content, you will rarely see this traffic and the decision hardly matters.
How do I write a rule that keeps matching?
Match the Claude-Code robots.txt token, or the claude-code/ prefix in server rules. The version number changes with every CLI release, so a rule pinned to a full user-agent string will stop matching.
Can Claude-Code read your page right now?
Test any URL and see exactly what AI crawlers receive.