OmniMCP

# OmniMCP > [!IMPORTANT] > **Labs, not the OpenAdapt product.** This repository is a perceive-plan-act > computer-use experiment. It is not how OpenAdapt records, compiles, or > replays a workflow. The product is > [openadapt-flow](https://github.com/OpenAdaptAI/openadapt-flow): a compiled > program that reports `VERIFIED` only if an independent check agrees. > `pip install openadapt`. [](https://github.com/OpenAdaptAI/OmniMCP/actions/workflows/ci.yml) [](https://opensource.org/licenses/MIT) [](https://www.python.org/) [](https://github.com/astral-sh/ruff) OmniMCP provides rich UI context and interaction capabilities to AI models through [Model Context Protocol (MCP)](https://github.com/modelcontextprotocol) and [microsoft/OmniParser](https://github.com/microsoft/OmniParser). It focuses on enabling deep understanding of user interfaces through visual analysis, structured planning, and precise interaction execution. ## Core Features - **Visual Perception:** Understands UI elements using OmniParser. - **LLM Planning:** Plans next actions based on goal, history, and visual state. - **Agent Executor:** Orchestrates the perceive-plan-act loop (`omnimcp/agent_executor.py`). - **Action Execution:** Controls mouse/keyboard via `pynput` (`omnimcp/input.py`). - **CLI Interface:** Simple entry point (`cli.py`) for running tasks. - **Auto-Deployment:** Optional OmniParser server deployment to AWS EC2 with auto-shutdown. - **Debugging:** Generates timestamped visual logs per step. ## Overview `cli.py` uses `AgentExecutor` to run a perceive-plan-act loop. It captures the screen (`VisualState`), plans using an LLM (`core.plan_action_for_ui`), and executes actions (`InputController`). ### Demos - **Real Action (Calculator):** `python cli.py` opens Calculator and computes 5*9.  - **Synthetic UI (Login):** `python demo_synthetic.py` uses generated images (no real I/O). *(Note: Pending refactor to use AgentExecutor).*  ## Prerequisites - Python >=3.10, <3.13 - `uv` installed (`pip install uv`) - **Linux Runtime Requirement:** Requires an active graphical session (X11/Wayland) for `pynput`. May need system libraries (`libx11-dev`, etc.) - see `pynput` docs. *(macOS display scaling dependencies are handled automatically during installation).* ### For AWS Deployment Features Requires AWS credentials in `.env` (see `.env.example`). **Warning:** Creates AWS resources (EC2, Lambda, etc.) incurring costs. Use `python -m omnimcp.omniparser.server stop` to clean up. ```.env AWS_ACCESS_KEY_ID=YOUR_ACCESS_KEY AWS_SECRET_ACCESS_KEY=YOUR_SECRET_KEY ANTHROPIC_API_KEY=YOUR_ANTHROPIC_KEY # OMNIPARSER_URL=http://... # Optional: Skip auto-deploy ``` ## Installation ```bash git clone [https://github.com/OpenAdaptAI/OmniMCP.git](https://github.com/OpenAdaptAI/OmniMCP.git) cd OmniMCP ./install.sh # Creates .venv, installs deps incl. test extras cp .env.example .env # Edit .env with your keys # Activate: source .venv/bin/activate (Linux/macOS) or relevant Windows command ``` ## Quick Start Ensure environment is activated and `.env` is configured. ```bash # Run default goal (Calculator task) python cli.py # Run custom goal python cli.py --goal "Your goal here" # See options python cli.py --help ``` Debug outputs are saved in `runs/<timestamp>/`. **Note on MCP Server:** An experimental MCP server (`OmniMCP` class in `omnimcp/mcp_server.py`) exists but is separate from the primary `cli.py`/`AgentExecutor` workflow. ## Architecture 1. **CLI** (`cli.py`) - Entry point, setup, starts Executor. 2. **Agent Executor** (`omnimcp/agent_executor.py`) - Orchestrates loop, manages state/artifacts. 3. **Visual State Manager** (`omnimcp/visual_state.py`) - Perception (screenshot, calls parser). 4. **OmniParser Client & Deploy** (`omnimcp/omniparser/`) - Manages OmniParser server communication/deployment. 5. **LLM Planner** (`omnimcp/core.py`) - Generates action plan. 6. **Input Controller** (`omnimcp/input.py`) - Executes actions (mouse/keyboard). 7. **(Optional) MCP Server** (`omnimcp/mcp_server.py`) - Experimental MCP interface. ## Development ### Environment Setup & Checks ```bash # Setup (if not done): ./install.sh # Activate env: source .venv/bin/activate (or similar) # Format/Lint: uv run ruff format . && uv run ruff check . --fix # Run tests: uv run pytest tests/ ``` ### Debug Support Running `python cli.py` saves timestamped runs in `runs/`, including: * `step_N_state_raw.png` * `step_N_state_parsed.png` (with element boxes) * `step_N_action_highlight.png` (with action highlight) * `final_state.png` Detailed logs are in `logs/run_YYYY-MM-DD_HH-mm-ss.log` (`LOG_LEVEL=DEBUG` in `.env` recommended). <details> <summary>Example Log Snippet (Auto-Deploy + Agent Step)</summary> ```log # --- Initialization & Auto-Deploy --- 2025-MM-DD HH:MM:SS | INFO | omnimcp.omniparser.client:... - No server_url provided, attempting discovery/deployment... 2025-MM-DD HH:MM:SS | INFO | omnimcp.omniparser.server:... - Creating new EC2 instance... 2025-MM-DD HH:MM:SS | SUCCESS | omnimcp.omniparser.server:... - Instance i-... is running. Public IP: ... 2025-MM-DD HH:MM:SS | INFO | omnimcp.omniparser.server:... - Setting up auto-shutdown infrastructure... 2025-MM-DD HH:MM:SS | SUCCESS | omnimcp.omniparser.server:... - Auto-shutdown infrastructure setup completed... ... (SSH connection, Docker setup) ... 2025-MM-DD HH:MM:SS | SUCCESS | omnimcp.omniparser.client:... - Auto-deployment successful. Server URL: http://... ... (Agent Executor Init) ... # --- Agent Execution Loop Example Step --- 2025-MM-DD HH:MM:SS | INFO | omnimcp.agent_executor:run:... - --- Step N/10 --- 2025-MM-DD HH:MM:SS | DEBUG | omnimcp.agent_executor:run:... - Perceiving current screen state... 2025-MM-DD HH:MM:SS | INFO | omnimcp.visual_state:update:... - VisualState update complete. Found X elements. Took Y.YYs. 2025-MM-DD HH:MM:SS | INFO | omnimcp.agent_executor:run:... - Perceived state with X elements. ... (Save artifacts) ... 2025-MM-DD HH:MM:SS | DEBUG | omnimcp.agent_executor:run:... - Planning next action... ... (LLM Call) ... 2025-MM-DD HH:MM:SS | INFO | omnimcp.agent_executor:run:... - LLM Plan: Action=..., TargetID=..., GoalComplete=False 2025-MM-DD HH:MM:SS | DEBUG | omnimcp.agent_executor:run:... - Added to history: Step N: Planned action ... 2025-MM-DD HH:MM:SS | INFO | omnimcp.agent_executor:run:... - Executing action: ... 2025-MM-DD HH:MM:SS | SUCCESS | omnimcp.agent_executor:run:... - Action executed successfully. 2025-MM-DD HH:MM:SS | DEBUG | omnimcp.agent_executor:run:... - Step N duration: Z.ZZs ... (Loop continues or finishes) ... ``` *(Note: Details like timings, counts, IPs, instance IDs, and specific plans will vary)* </details> ## Roadmap & Limitations Key limitations & future work areas: * **Performance:** Reduce OmniParser latency (explore local models, caching, etc.) and optimize state management (avoid full re-parse). * **Robustness:** Improve LLM planning reliability (prompts, techniques like ReAct), add action verification/error recovery, enhance element targeting. * **Target API/Architecture:** Evolve towards a higher-level declarative API (e.g., `@omni.publish` style) and potentially integrate loop logic with the experimental MCP Server (`OmniMCP` class). * **Consistency:** Refactor `demo_synthetic.py` to use `AgentExecutor`. * **Features:** Expand action space (drag/drop, hover). * **Testing:** Add E2E tests, broaden cross-platform validation, define evaluation metrics. * **Research:** Explore fine-tuning, process graphs (RAG), framework integration. ## Project Status Core loop via `cli.py`/`AgentExecutor` is functional for basic tasks. Performance and robustness need significant improvement. MCP integration is experimental. ## Contributing 1. Fork repository 2. Create feature branch 3. Implement changes & add tests 4. Ensure checks pass (`uv run ruff format .`, `uv run ruff check . --fix`, `uv run pytest tests/`) 5. Submit pull request ## License MIT License ## Contact - Issues: [GitHub Issues](https://github.com/OpenAdaptAI/OmniMCP/issues) - Questions: [Discussions](https://github.com/OpenAdaptAI/OmniMCP/discussions) - Security: security@openadapt.ai