Skip to content

Say on the MCP start card that a server can steer the Bot - #1623

Merged
DoodleBears merged 1 commit into
mainfrom
claude/mcp-untrusted-content
Oct 14, 2026
Merged

DoodleBears merged 1 commit into
mainfrom
claude/mcp-untrusted-content

Conversation

@DoodleBears

@DoodleBears DoodleBears commented Oct 13, 2026 •

Copy link
Copy Markdown
Contributor

Requested by Vain

Why the change

An MCP server can add its own instructions to the Bot's system prompt, and its tool descriptions and results reach the model. The card for starting a server did not say so.

Closes #1620

Special things to note

  • Merge risk: text only. Revert by reverting this PR.
  • Following common MCP client practice, the Human decides which servers to trust and BotHarness states the risk; it does not filter or relabel what a server sends. The start approval card now says the server can add its own instructions to the Bot's system prompt and that its tool descriptions and results can influence the Bot, and to connect only a trusted server. The ordinary Profile guide says the same next to the MCP steps, in both languages.
  • An earlier version of this branch also dropped server instructions in the Host's launcher and added a prompt statement. Measured on fresh Bots with a test server carrying injected text: main 0 of 4 followed it; with only the statement 1 of 4 followed the server's instructions; with instructions dropped 0 of 8. The owner chose not to filter; that version was replaced by this one and the measurements are kept in the issue.
  • Verification: pnpm lint, pnpm format:check, pnpm typecheck pass; the approval card and settings component tests pass. No behaviour changed, so no runtime run. Human QA: not done.

Change outline

Only the warning sentence on the MCP start approval card changes on screen.

🤖 Generated with Claude Code

@DoodleBears DoodleBears self-assigned this Oct 13, 2026
@coderabbitai

coderabbitai Bot commented Oct 13, 2026 •

Copy link
Copy Markdown

Warning

Review limit reached

You've used all free OSS reviews for now. Wait for the free limit to reset to keep reviewing this public repository.

Next included review available in 37 minutes.

Check out review usage here.

View limit details

Limit details: You’ve used the included review currently available.

Learn how review limits work.

Review configuration:

⚙️ Run configuration
  • Configuration used: defaults
  • Review profile: CHILL
  • Plan: Advanced
  • Run ID: 22653bad-8a7a-44d2-9108-447b1e43efd8
📥 Commits

Reviewing files that changed from the base of the PR and between 7fe5079 and 068f16f.

📒 Files selected for processing: 5
  • CHANGELOG.md
  • CHANGELOG.zh.md
  • docs/dev/guides/private-input-ordinary.md
  • docs/dev/guides/private-input-ordinary.zh.md
  • packages/client/src/client/locale.ts
  • Autofix · Keep fixing CodeRabbit findings and required CI, and resolving merge conflicts

Thanks for using CodeRabbit! It's free for OSS, and your support helps us grow. If you like it, consider giving us a shout-out.

❤️ Share

Comment @coderabbitai help to get the list of available commands.

…e Bot

Issue: #1620
Agent-Task: claude-code/local/9f51f103-7177-4bd0-8fd7-81d1ce39b98f

Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
@DoodleBears
DoodleBears force-pushed the claude/mcp-untrusted-content branch from 3dc90d5 to 068f16f Compare October 14, 2026 05:10
@DoodleBears DoodleBears changed the title Keep MCP server instructions out of the system prompt and name server content as outside data Say on the MCP start card that a server can steer the Bot Oct 14, 2026
@DoodleBears
DoodleBears marked this pull request as ready for review October 14, 2026 05:32
@DoodleBears
DoodleBears merged commit d10c2da into main Oct 14, 2026
2 checks passed
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

Say on the MCP start card that a server can steer the Bot

1 participant