[FEATURE] One switch to turn off Claude Docs in Claude Code (today it arrives as a "Custom" connector + a separate skill)

Preflight Checklist

• [x] I have searched existing requests and this feature hasn't been requested yet
• [x] This is a single feature request (not multiple features)

Problem Statement

Claude Docs is a great addition to claude.ai. But in Claude Code it arrives in a way that's hard to understand and impossible to turn off cleanly.

What a user sees:

• A connector they didn't add. Claude Desktop → Settings → Connectors lists "Claude Docs" as Web · Custom, looking the same as connectors I added myself (Context7, DeepWiki). In the terminal it appears in /mcp as claude.ai Claude Docs. Yet the support article says Claude Docs isn't a connector, and that individual accounts can't disable it.
• A skill they didn't add. Settings → Skills shows a docs skill "by Anthropic" with its own toggle. Its description is mostly routing rules, and it tells the model to call the connector's guide tool if the connector's instructions are missing.
• The same guidance, three times. The connector's instructions, the guide tool (which re-serves those instructions) and the skill's description all carry overlapping guidance. Turning off the skill leaves the rest in place.

Opting out depends on where you are:

| Surface | How to turn it off |
|---|---|
| Terminal CLI | Only via "deniedMcpServers": [{ "serverName": "claude.ai Claude Docs" }], a general MCP block, not a feature setting |
| Desktop app, Code tab | Not possible. Desktop delivers connectors in-process, so disableClaudeAiConnectors and deniedMcpServers don't apply (MCP docs), and Claude Docs can't be disconnected at claude.ai like other connectors |

Why it matters: Anthropic created MCP and sets the example for how tools, skills and connectors should fit together, and Claude Code usually gets this right. Claude Docs doesn't follow that example: one first-party feature is spread across a connector, a skill, a tool and always-on instructions, each managed somewhere different. For people who care about keeping their context, connectors and plugins clean and understood, finding a "Custom" connector and a skill they never added is confusing. It also makes them wonder what else is loaded that they can't see or control.

Proposed Solution

One user-level switch for Claude Docs in Claude Code, covering the whole feature (connector, instructions and skill) in both the terminal CLI and the Desktop Code tab. Ideally it would be separate from claude.ai chat, so it can stay on in chat, where it's most useful, and be off in a coding tool.

Related to that, show it as a built-in feature: labeled as built in rather than "Custom", and appearing once rather than as a connector plus a skill.

Alternative Solutions

• CLI: deniedMcpServers works, but it's a workaround and doesn't reach Desktop.
• Desktop: turning off the docs skill leaves the connector and its instructions loaded, and switching the connector off from + → Connectors only lasts for the current session.

Priority

Low - Nice to have

Feature Category

MCP server integration

Use Case Example

1. I use Claude Code, in the terminal and in the Desktop Code tab, only for coding.
2. I keep connectors minimal on purpose (disableClaudeAiConnectors, deniedMcpServers).
3. After Claude Docs launched, a "Custom" connector and an Anthropic skill appeared that I never added, loaded into every session.
4. With one switch, I'd turn Claude Docs off for Claude Code, keep it in claude.ai chat, and know exactly what's in my context.

Additional Context

Environment: Claude Desktop 2.9939.2 · Claude Code 2.1.282 · macOS · Max plan

Screenshots in the first comment: Settings → Connectors (Claude Docs shown as "Web · Custom") and the docs skill page in Settings → Skills.