Return a logger with *name* (defaults to ``anytool``). The first call triggers :meth:`configure` automatically.

Write to stderr, filtering out harmless error messages. Args: s: The string to write Returns:

Load cache from disk. Returns empty dict if not exists.

Send a JSON-RPC request and return the result. Args: method: The JSON-RPC method name (e.g., "tools/list", "tools/call")

Exit the async context manager. Args: exc_type: The exception type, if an exception was raised. exc_val: The exceptio

Get global configuration instance. Usage: - Get configuration in Provider: get_config().get_backend_config('shell') - Ge

Record one step operation Args: backend: backend type (gui/shell/mcp/web/system) tool: tool name (na

Save tool metadata to disk (overwrites existing cache). Args: servers: Dict mapping server_name -> list of tool

Ensure the connector is connected.

Clear all cached data.

Close the HTTP session.

Single-round LLM call with optional tool execution. Args: messages: conversation history (List[Dict] for standar

Flush any queued telemetry events

Get trajectory directory path

Ensure the connector is connected, reconnecting if necessary. Raises: RuntimeError: If connection cannot be established and auto_

Add log message

List tools from backend(s) or session. 1. session_name is provided → return tools from that session 2. backend is li

Safely parse tool_call.function.arguments which may be JSON string. Handles: 1. Proper JSON strings with true/false/null 2. P

Send a request and wait for a response.

Bind runtime information to the tool instance. Allow the tool to be invoked directly without specifying backend/session/server.

Flush any remaining buffered content and the underlying stream.

Set self._active to False

Compute hash of tool description for change detection.

Internal helper to get mcpServers configuration. Tries both 'mcpServers' and 'servers' keys for compatibility. Retur

Generate platform-specific conda activation command prefix Args: conda_env: Conda environment name (e.g., 'myenv') Retu

Return all configured MCP server names from MCPClient config. Returns: List of server names

Load agent actions from a trajectory directory.

Set self._active to True

Ensure cache directory exists.

Extract package name from npx arguments. Args: args: npx arguments list, e.g. ["-y", "mcp-excalidraw-server"] or ["bazi-m

Fetch tools from provider. Args: backend: Backend type session_name: - None: fetch

Record tool scores from ranking results.

Execute operation with retry logic. Args: operation_name: Name of operation for logging operation_fu

Save metadata to metadata.json

Capture a telemetry event

Create a new MCP session for a specific server. Args: session_config: Must contain 'server' in connection_params

End screen recording and optionally save to file.

Get or create quality record for a tool.

Get global tool cache instance.

Generate unique key for a tool.

Initialize provider, call `create_session` to create all sessions if not exist Subclasses should set `self.is_initialized = True` afte

Invoke this tool using bound runtime information. Requires runtime info to be bound via bind_runtime_info(). If no runtime in

Load complete recording session including trajectory, metadata, plans, and snapshots. Args: recording_dir: Path to recording dir

Stop recording screen and save to local

Extract package name from uvx arguments. Args: args: uvx arguments list, e.g. ["--from", "office-powerpoint-mcp-server",

Extract package name from uv run arguments. Args: args: uv arguments list, e.g. ["run", "--with", "biomcp-python", "biomc

Get or create aiohttp session.

Check if the collected traceback is a harmless error.

Load single JSON configuration file. This function wraps the generic load_json_file and adds global configuration specific error handling and

Parse tool data into Tool objects.

Rebuild text-to-key mapping for quick lookup.

Run a command via asyncio subprocess and return a result dict matching the format returned by the local_server endpoints.

Capture screenshot via HTTP API. Calls: GET {base_url}/screenshot Returns: PNG image bytes, or

return True if allowed, False if denied. Command check uses token-level matching to prevent simple space/escape bypasses.

Close all sessions and release resources. Automatically called when using context manager.

Close an MCP session by session name. Args: session_name: Session name in format 'mcp-<server_name>'

Configure the logging system. Usually called automatically on first use; pass ``force=True`` to reconfigure explicitly. Args

Disconnect from GUI desktop environment

Internal helper. Guarantee that `initialize()` has been executed

Execute a Python command on desktop environment. Used for pyautogui commands. Args: command: Python comm

Get configuration for specified backend

Get base URL for connecting to local server. Priority: 1. Environment variable LOCAL_SERVER_URL 2. Read from local_server/config

Read local server configuration. Priority: 1. Environment variable LOCAL_SERVER_URL (parsed into host/port) 2. Config file local

Get the current mcp-use package version.

Get current desktop screenshot.

Get comprehensive system information. Returns information including: - system: OS name (Linux, Darwin, Windows)

Negotiate with the backend, discover tools, etc. Return session information (can be an empty dict). `self.tools` nee