Notify dashboard that an MCP App launched — embed as panel.

Run a headless chat loop for a spawned agent. Similar to ``_run_enhanced_chat_loop`` in ``chat_handler.py`` but: * No terminal I/O — reads p

Save current session to disk. Returns: Path to saved file, or None on failure.

Determine if a tool should be confirmed based on preferences. Args: tool_name: Name of the tool Returns: Tru

Switch to a specific provider and model.

Validate server configuration and return status and errors. Args: cfg: MCPConfig instance servers: List of server names to v

Build activity stream events from conversation history. Pairs assistant ``tool_calls`` with their corresponding ``tool``-role result

Extract a tool call from an LLM completion response. Handles three response formats: 1. chuk_llm native: {"response": ..., "tool_calls": [...

Get a client from chuk_llm for standard providers. Args: provider: Provider name model: Model name (optional

Initialize StreamManager with detected transport type. ENHANCED: Initializes different transport types in parallel for faster startup.

Return a mock ChatContext.

Await a result if it's a coroutine, otherwise return it directly. Allows callbacks to be either sync or async functions.

Resolve ${var} references in tool arguments. Supports: - ${variable} — direct replacement - ${variable.field} — nested dict access -

Close the ToolManager, swallowing any exception during shutdown.

Finish tool execution display. Args: result: Tool execution result success: Whether execution succeeded

Get list of available providers.

Get environment variable value (type-safe). Args: var: EnvVar enum member default: Default value if not set Returns:

Get environment variable as boolean. Args: var: EnvVar enum member default: Default value if not set Returns: Boolea

Determine server icon based on MCP capabilities. Args: capabilities: Server capabilities dictionary tool_count: Number of to

Get current theme.

Execute a memory tool and return the result as a string.

Initialize or get the configuration. Priority order: 1. Explicit config_path if provided 2. server_config.json in cu

Convenience function to initialize the configuration. Args: config_path: Path to configuration file Returns: The initia

Interrupt current streaming operation.

Check if a server URL belongs to a trusted domain. Args: server_url: The server URL to check (may be None for STDIO servers)

Load a saved session into the current context. Args: session_id: Session ID to load Returns: True if loaded

Log error message.

Synchronous convenience wrapper (used by `mcp-cli` entry-point and tests). Spins up its own event-loop when necessary.

Set confirmation for a specific tool. Args: tool_name: Name of the tool setting: 'always', 'never', 'ask', or None (r

Start tool execution display. Args: tool_name: Name of tool being executed arguments: Tool arguments

Convert to dict.

Update an existing custom provider. Args: name: Provider name to update api_base: New API base URL (optional)

Validate parameters before execution. Returns: Error message if validation fails, None if valid.

Build the server/tool categorization section for the system prompt. Args: server_tool_groups: Typed list of server-to-tools groupings.

Map MIME type to a UI kind: image, text, audio, or unknown.

Extract tool name, arguments, and call ID from a tool call.

Handle streaming completion with UI integration. Args: tools: Tool definitions to pass to the LLM, or None to disable tools

Execute a VM tool (page_fault or search_pages) via MemoryManager. VM tools are internal memory operations that bypass the MCP ToolManager

Check if an error message indicates OAuth authorization is needed.

Parse raw arguments into a dictionary.

Record token usage from a completion into the tracker.

Check if tool requires user confirmation.

Convert an object to a JSON-serializable form.

Validate a plan step against available tools.

Filter events to fit within VM token budget. Groups events into logical turns (starting at each user message), estimates token cost p

Public broadcast method for external callers (e.g. commands).

Clear the staging area.

Create OAuth token refresh callback for StreamManager. Args: http_servers: List of HTTP server configs sse_servers: L

Delete a saved session. Args: session_id: Session ID to delete Returns: True if deleted, False if not found

Detect which servers are HTTP vs STDIO based on configuration. Args: cfg: MCPConfig instance servers: List of server names t

Find a tool by its name (handles both simple and namespaced names). Uses O(1) dict lookup instead of linear scanning.

Format server capabilities as readable string. Args: capabilities: Capabilities dictionary from server Returns: Formatt

Format content for inline preview. Args: content: Full content max_len: Maximum length to show Returns: Formatted pr

Convenience function to get the current configuration. Returns: The current MCPConfig Raises: RuntimeError: If config h

Get the current configuration. Raises: RuntimeError: If config hasn't been initialized

Record a turn's token usage.

Remove a custom provider. Args: name: Provider name to remove Returns: True if provider was removed, False i

Resolve effective provider and model from user input. Args: provider: User-specified provider (optional) mod

Save a named layout to disk, replacing any existing layout with the same name.

Register the asyncio.Queue that the chat loop reads from. Browser messages (USER_MESSAGE / USER_COMMAND) are put on this queue so the

Find an available port and start the server. Returns the bound port.

Stop streaming response.

Switch to a different provider and its default model.

Validate a single tool for LLM compatibility.

Build CONFIG_STATE payload from current ChatContext.

Build conversation history payload from ChatContext. Returns only user/assistant messages — tool-role messages are excluded from the

Check if a tool has an MCP Apps UI and launch/update it. Handles two cases per the MCP Apps spec: 1. Tool has resourceUri — reuse an

Signal UI that all tool calls are complete.

Get models from chuk_llm for a provider. Tries multiple strategies: 1. Get from chuk_llm's cached provider info 2. If only "*

Look up the server URL for a tool using cached context data.

Callback when a tool completes. ENHANCED: Now includes value binding system for dataflow tracking. - Binds numeric results to $vN ide

Extract the action string from kwargs/args.

Parse a single tool call entry from either dict or object format. Args: tc: A tool call entry (dict or object with function attribute).

Receive one message with a timeout.

Start background health polling if configured.

Keep reasoning_content only on the most recent assistant message. Thinking models (DeepSeek, Kimi) produce 100K+ chars of reasoning per

Try to parse a string as JSON. Returns parsed value or original string.

Build a dashboard-safe descriptor for an Attachment. Keeps payloads small by applying thresholds: - Images < *inline_threshold* bytes: includ

Find image URLs in message text.

Detect MIME type from file extension.