Return OpenAI-format tool definitions for memory scope tools.

Get provider API key using hierarchical resolution. Hierarchy: 1. Environment variable (highest priority) 2. Token storage (via Toke

Get a specific runtime server configuration.

Get all runtime servers.

Get the server name for a given tool.

Get active streams from StreamManager.

Get tool configuration value with priority resolution. Args: key: Configuration key (e.g., "include_tools", "confirm_too

Get the global tool confirmation mode.

Determine the risk level of a tool based on patterns. Args: tool_name: Name of the tool Returns: Risk level

Initialize or get the application context.

Initialize tools, session, and procedural memory. Args: on_progress: Optional callback for progress updates during startup

Stub for removed inject_logging_env_vars function.

Inject a tool-related message into conversation history. Tool messages (assistant with tool_calls, tool results) have special struct

Check if a server is a runtime server.

List all saved plans as dicts. Returns: List of plan summaries with id, title, step count.

List available prompts from servers.

List available resources from servers.

List all saved sessions. Returns: List of session metadata, sorted by updated_at (newest first)

Synchronous load (for backward compat during transition).

Error logging with rich formatting if available.

Broadcast a plan update to the dashboard.

Register all built-in commands with the unified registry. This should be called once during application startup.

Initialise the ToolManager and context, then call *async_command(...)*. The *async_command* may itself be `async` **or** synchronous – both

Register a callback to execute a browser-requested tool call. The callback receives (tool_name, arguments) and should execute the too

Stop all managed agents.

Add cancelled tool call to conversation history.

Build the system prompt for LLM plan generation. Reuses the same logic as plan.py's _build_plan_system_prompt.

Recursively check for array schemas missing 'items' property.

Run mcp-cli's guard checks before tool execution. Returns an error message if blocked, None if allowed.

Run health check on a server and return a diagnostic string.

Display a plan in the terminal with DAG visualization.

Extract a human-readable error message from an MCP error result.

Unwrap middleware/ToolCallResult wrappers to get the raw MCP result.

Fetch models from provider's /models API endpoint. Works for OpenAI-compatible APIs (deepseek, openai, etc.)

Resolve tool timeout for a server: per-server → global → default. Args: server_name: Server namespace/name, or None for global de

Get tool list from CLI/env (type-safe!).

Handle approval/denial response from the dashboard.

Immediate interrupt (compatibility method).

Check if an error message indicates a connection problem.

Create a minimal ChatContext-like object with _tool_index populated.

Create a mock ModelManager with a sequence of LLM responses. Each response can have 'tool_calls' (LLM wants to call a tool) or just '

Build a RuntimeConfig mock with distinct timeout values.

Create a fake ToolInfo-like object.

Parse a string like "content=[{'type': 'text', 'text': '4.2426'}]".

Extract (tool_name, arguments_str, call_id) from a tool_call.

Make variables JSON-serializable. Truncates large values to prevent bloated checkpoint files.

Show available slash commands when just '/' is typed.

Spawn a mock agent, returns the manager.

Stop background health polling.

Build a compact summary of current variables for the LLM.

Process and display a new chunk. Args: raw_chunk: Raw chunk data from LLM provider

Add models to the provider and update default if needed. Args: models: List of model IDs to add

Increment auto-save counter and save when threshold is reached.

Thin wrapper so `uv run mcp-cli chat` (or `interactive`) is minimal.

Close a specific app and its server.

Factory for test contexts.

Delete a plan by ID (or unique prefix). Args: plan_id: The plan's UUID or a unique prefix. Returns: True if

Filter search results based on state. Parameterized tools (normal_cdf, etc.) are blocked/downranked until computed values exist in st

Find a tool by name.

Fix OpenAI compatibility issues by removing unsupported properties. Returns: Fixed tool definition

Return OpenAI-format tool definitions for agent orchestration.

Get a bridge for any running app (preferring READY state). Used to route ui_patch results from tools that don't carry a resourceUri t

Get tool confirmation setting (legacy compatibility).

Get resolved tool configuration.

Get the list of trusted domains.

Get user input with async prompt. Args: prompt: Prompt text to display Returns: User input string

Return status dicts for all managed agents.

Log success message.

Called for each streamed LLM token (high-volume — only used by agent terminal).

Build an ``Attachment`` from browser-uploaded file data. Parameters ---------- filename: Original filename from the browser ``Fil

Register all CLI-compatible commands with a Typer app. Args: app: The Typer application to register commands with.

Render tool execution status with argument preview. Args: tool: Tool execution state spinner: Current spinner frame elaps

Enable or disable auto-fixing.

Display a user message. Args: message: User message content

Test an mcp-cli command and return success status and output.

Convert to dict for LLM API calls. Handles provider-specific requirements: - OpenAI: Requires 'content' field in assistant messages w

Format capabilities as readable string.

Trigger discovery after environment setup. Returns: Number of new functions discovered

Create a user message.

Comprehensive tool validation and fixing. Returns: Tuple of (is_valid, fixed_tool_def, error_message)

Validate if a model exists for a provider. Args: model: Model name to validate provider: Provider name (uses

Validate if a provider exists. Args: provider: Provider name to validate Returns: True if provider