[Bug] MCP Server Detection Fails with --dangerously-skip-permissions
Bug Description
When a repository contains a .mcp.json file with MCP server configurations, Claude Code does not automatically detect or use these servers when running with the --dangerously-skip-permissions flag, even though this flag is meant to bypass permission checks.
Steps to Reproduce
- Have a repository with a valid .mcp.json file in the root (e.g., configuring a gopls MCP server)
- Start Claude Code with claude --dangerously-skip-permissions
- Run claude mcp list
- Observe: "No MCP servers configured" message appears, project MCP servers are not detected
Expected Behavior
When using --dangerously-skip-permissions, project-level MCP configurations should be automatically trusted and loaded without requiring user interaction.
Actual Behavior
Project MCP servers are only recognized after:
- Running Claude Code once WITHOUT the --dangerously-skip-permissions flag
- Explicitly trusting the MCP servers when prompted
- Then subsequent runs with --dangerously-skip-permissions work correctly
Workaround
Start Claude Code once without --dangerously-skip-permissions to trust the project's MCP servers, then the flag works as expected in future sessions.
Impact
This creates confusion for users who expect --dangerously-skip-permissions to bypass all permission checks, and makes automated/CI workflows more difficult since they require an initial manual trust step.
Environment Info
- Platform: linux
- Terminal: windows-terminal
- Version: 1.0.98
- Feedback ID:
Errors
[{"error":"Error: Request was aborted.\n at d71._createMessage (file:///home/fank/.nvm/versions/node/v24.1.0/lib/node_modules/@anthropic-ai/claude-code/cli.js:970:7955)\n at process.processTicksAndRejections (node:internal/process/task_queues:105:5)","timestamp":"2025-08-30T06:27:10.913Z"},{"error":"AggregateError\n at sc.from (file:///home/fank/.nvm/versions/node/v24.1.0/lib/node_modules/@anthropic-ai/claude-code/cli.js:671:39831)\n at zK.I1 (file:///home/fank/.nvm/versions/node/v24.1.0/lib/node_modules/@anthropic-ai/claude-code/cli.js:674:10122)\n at zK.emit (node:events:507:28)\n at zK.emit (node:domain:489:12)\n at mm1.<computed> (file:///home/fank/.nvm/versions/node/v24.1.0/lib/node_modules/@anthropic-ai/claude-code/cli.js:25:1161)\n at ClientRequest.emit (node:events:507:28)\n at ClientRequest.emit (node:domain:489:12)\n at emitErrorEvent (node:_http_client:104:11)\n at TLSSocket.socketErrorListener (node:_http_client:518:5)\n at TLSSocket.emit (node:events:507:28)\n at w21.request (file:///home/fank/.nvm/versions/node/v24.1.0/lib/node_modules/@anthropic-ai/claude-code/cli.js:676:2088)\n at process.processTicksAndRejections (node:internal/process/task_queues:105:5)\n at async oF5 (file:///home/fank/.nvm/versions/node/v24.1.0/lib/node_modules/@anthropic-ai/claude-code/cli.js:2429:470)\n at async Promise.all (index 0)\n at async file:///home/fank/.nvm/versions/node/v24.1.0/lib/node_modules/@anthropic-ai/claude-code/cli.js:2415:9180","timestamp":"2025-08-30T06:34:57.069Z"}]This issue has 3 comments on GitHub. Read the full discussion on GitHub ↗