Brummer MCP Server

# Agentic AI Coders Feature (Tmux-Style Design) ## Overview Brummer will support running AI coding assistants as interactive sessions within the TUI, similar to tmux sessions. This creates a seamless development experience where AI coders receive real-time feedback from Brummer's build system, test runners, and error detection. ## Core Concepts ### Interactive AI Coder Sessions - **Tmux-style attach/detach**: AI coders run in persistent sessions that users can attach to and detach from - **Direct interaction**: Users can type commands and see AI responses in real-time within the Brummer TUI - **Session persistence**: AI coders continue working even when detached, similar to tmux sessions - **Multiple sessions**: Support for running multiple AI coders simultaneously with easy switching ### Real-Time Brummer Integration - **Error reporting**: Build errors, test failures, and lint warnings are automatically sent to the AI - **Context awareness**: AI receives file changes, process outputs, and system events - **Proactive assistance**: AI can suggest fixes before the user even asks - **Workspace isolation**: Each AI coder operates in its own sandboxed workspace ## User Interaction Model ### Starting an AI Coder ```bash # Via TUI Press 'a' to switch to AI Coders view Press 'n' to create new AI coder Select provider (Claude, GPT-4, etc.) Enter initial task/prompt # Via MCP ai_coder_create {"task": "implement user auth", "provider": "claude"} ``` ### Interacting with AI Coders ``` ┌─────────────────────────────────────────────────────────────┐ │ AI Coder: implement-auth (Claude) - Running │ ├─────────────────────────────────────────────────────────────┤ │ [AI] I'll implement user authentication. Starting with... │ │ │ │ [Brummer] Build error in auth.go:15 │ │ undefined: bcrypt.GenerateFromPassword │ │ │ │ [AI] I see the issue. We need to import bcrypt. Let me... │ │ │ │ [User] > also add JWT token generation │ │ │ │ [AI] Sure! I'll add JWT token generation to the auth... │ ├─────────────────────────────────────────────────────────────┤ │ Progress: ████████░░ 80% | Files: 5 | Tokens: 12.5k │ └─────────────────────────────────────────────────────────────┘ ``` ### Session Management Commands - `Ctrl+b d` - Detach from current AI coder (continues running) - `Ctrl+b n` - Next AI coder session - `Ctrl+b p` - Previous AI coder session - `Ctrl+b w` - List all AI coder sessions - `Ctrl+b c` - Create new AI coder - `Ctrl+b &` - Kill current AI coder session ## Brummer Hooks and Events ### Automatic Error Reporting When Brummer detects errors, they're automatically sent to attached AI coders: ```go // Build error detected Brummer → AI: "Build failed: undefined variable 'user' at handlers.go:45" AI → User: "I see there's an undefined variable. Let me check the context..." // Test failure Brummer → AI: "Test failed: TestUserLogin - expected 200, got 401" AI → User: "The login test is failing. This might be due to..." // Lint warning Brummer → AI: "Lint: exported function CreateUser should have comment" AI → User: "I'll add the missing documentation comment..." ``` ### Event Hooks Integration ```yaml AI Coder Hooks: - process_failed: Send stdout/stderr to AI - build_error: Send error details and file context - test_failed: Send test output and failure reason - lint_warning: Send lint messages - file_changed: Notify AI of external file modifications ``` ## Implementation Architecture ### Component Structure ``` /internal/aicoder/ ├── manager.go # AI coder lifecycle management ├── session.go # Interactive session handling (PTY) ├── provider.go # AI provider interface (Claude, GPT-4, etc.) ├── workspace.go # Isolated workspace management └── hooks.go # Brummer event integration /internal/tui/ └── ai_coder_view.go # Tmux-style TUI view /internal/mcp/ └── ai_coder_tools.go # MCP tools for external control ``` ### Session Lifecycle 1. **Create**: Initialize AI coder with provider and task 2. **Attach**: Connect to PTY for interactive communication 3. **Process**: Handle user input and AI responses 4. **Hook Events**: Receive and process Brummer events 5. **Detach**: Disconnect while keeping session alive 6. **Reattach**: Reconnect to existing session 7. **Terminate**: Clean up resources and workspace ## Example Workflows ### Debugging with AI Assistance ``` User: [Runs tests, sees failure] Brummer → AI: "Test TestAPIAuth failed: token validation error" AI: "I see the token validation is failing. Let me check the JWT implementation..." AI: [Makes code changes] Brummer → AI: "Build successful, running tests..." Brummer → AI: "All tests passing" AI: "Great! The authentication is now working correctly." ``` ### Collaborative Feature Development ``` User: "Implement rate limiting for the API" AI: "I'll implement rate limiting. Which approach would you prefer?" User: "Use Redis-based sliding window" AI: [Starts implementation] Brummer → AI: "Build error: Redis client not found" AI: "We need to add the Redis dependency. Let me update go.mod..." ``` ## Configuration ```toml [aicoder] default_provider = "claude" max_concurrent = 3 workspace_dir = "~/.brummer/ai-workspaces" [aicoder.providers.claude] api_key_env = "CLAUDE_API_KEY" model = "claude-3-sonnet" max_tokens = 100000 [aicoder.hooks] enabled = true include_build_errors = true include_test_failures = true include_lint_warnings = true context_lines = 10 ``` ## Security and Isolation - **Workspace Isolation**: Each AI coder operates in a separate directory - **Resource Limits**: CPU, memory, and disk usage limits per AI coder - **File Access**: Restricted to workspace and project directories only - **Network Access**: Configurable restrictions on external API calls - **Code Execution**: Optional sandboxing for running generated code ## Future Enhancements See the [main Brummer Roadmap](/docs/ROADMAP.md) for comprehensive feature plans and timelines. ### Phase 3: AI Coders Enhancement (Q2 2025) - **Multi-AI Collaboration**: Multiple AI coders working on different parts of the same feature - **Knowledge Persistence**: AI coders maintaining context across sessions - **Custom Tools Integration**: Ability to give AI coders access to specific development tools ### Phase 4: Environment & Configuration Management (Q2-Q3 2025) - **Environment Variable Management**: Unified .env file management with TUI interface - **Secrets Management**: Encrypted environment variables - **AI Integration**: Suggest appropriate environment configurations ### Phase 6: Team Collaboration (Q4 2025) - **Session Sharing**: Share AI coder sessions with team members - **Collaborative Debugging**: Multiple developers on same session For detailed specifications and implementation plans, see [/docs/ROADMAP.md](/docs/ROADMAP.md).