Skip to content

docs: clarify plugin MCP tool permissions - #1141

Open
yaodong-shen wants to merge 1 commit into
anthropics:mainfrom
yaodong-shen:docs/448-plugin-mcp-permissions
Open

docs: clarify plugin MCP tool permissions#1141
yaodong-shen wants to merge 1 commit into
anthropics:mainfrom
yaodong-shen:docs/448-plugin-mcp-permissions

Conversation

@yaodong-shen

Copy link
Copy Markdown

Closes #448

Summary

  • Explain that plugin MCP servers start automatically, but their tools still use the standard MCP permission flow.
  • Document how allowed_tools, disallowed_tools, permission_mode, can_use_tool, and PreToolUse hooks affect plugin MCP calls.
  • Clarify that tool permission checks do not sandbox the plugin server process, so plugins must be trusted.
  • Keep the public SdkPluginConfig and ClaudeAgentOptions.plugins docstrings consistent with the README.

Validation

  • .venv/bin/ruff check src/ tests/ scripts/
  • .venv/bin/ruff format --check src/ tests/ scripts/
  • .venv/bin/mypy src/ scripts/
  • .venv/bin/pytest -q: 1279 passed, 5 skipped; four process-state checks were blocked because the restricted test environment could not run ps
  • .venv/bin/pytest -q tests/test_close_cancellation.py outside the restricted environment: 9 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.

Unclear Documentation: Do Plugin MCP Servers Automatically Receive Permissions?

1 participant