AI Agent Overview

This section is for Codex, Claude, Cursor, and similar agents configuring or operating this MCP server.

Goal

Get to a reliable end state where:

• MCP server binary is built
• Flutter app is instrumented with mcp_toolkit
• client MCP config is updated safely (without deleting other servers)
• tool calls work against the correct debug target

If you need deterministic terminal automation instead of MCP client integration, see CLI vs MCP.

Start Here

1. Follow the Execution playbook.
2. Apply client-specific config:
   • Codex
   • Cursor
   • Claude
3. Validate with AI troubleshooting.

Canonical Agent Prompt File

If your assistant can follow a remote install guide, use:

• llm_install.md (GitHub)

Agent Safety Rules

• Ask for absolute paths before cloning or editing user files.
• Merge JSON changes instead of replacing entire config files.
• Preserve existing MCP servers in user config.
• Verify get_extension_rpcs includes ext.mcp.toolkit.* before claiming screenshot/layout/error inspection works.
• When multiple debug targets exist, retry with arguments.connection.targetId.