Struct Agent 

pub struct Agent { /* private fields */ }

MAX_ITERATIONS Documentation:

Why 200 iterations?

• Sufficient for most common tasks (file edits, code review, simple builds)
• Prevents infinite loops and runaway operations
• Balances task completion with resource efficiency

What happens when limit is reached?

• Agent stops execution gracefully
• User receives detailed warning message explaining:
  • Task status (may not be complete)
  • Reason for stopping (iteration limit)
  • Next steps (continue, break down task, or resume)

Future improvements:

• Dynamic adjustment based on task complexity
• User-configurable limits in config file
• Auto-resume with state preservation
• Progress indicators showing iteration count

Examples:

• Simple task (edit file): ~5-10 iterations
• Medium task (refactor module): ~15-30 iterations
• Complex task (build system): ~40-50 iterations (may hit limit)

Full Agent with event output

Implementations

impl Agent

pub fn event_sender(&self) -> Sender<AgentEvent>

Get event sender for streaming

pub fn set_ask_channel(&mut self, rx: Receiver<String>)

Set ask response channel (for TUI mode)

pub fn set_proxy_executor( &mut self, executor: Arc<dyn ProxyToolExecutor>, tool_defs: Vec<ProxyToolDef>, )

设置代理工具执行器

pub fn set_cancel_token(&mut self, token: CancellationToken)

Set cancellation token

pub fn set_approve_mode(&mut self, mode: ApproveMode)

Set approve mode at runtime

pub fn approve_mode_shared(&self) -> Arc<AtomicU8>

Get a shared reference to the approve mode atomic.

pub fn set_approve_mode_shared(&mut self, shared: Arc<AtomicU8>)

Replace the internal approve mode with an externally-created shared atomic.

pub fn update_memory_summary(&mut self, summary: Option<String>)

Update memory summary and rebuild system prompt. Note: Uses build_system_prompt (without project_path) to preserve cache.

pub fn refresh_codegraph_tools(&mut self)

Refresh CodeGraph tools after /init or codegraph init. This rebuilds both tools and system prompt with project_path. Call this only when CodeGraph state changes (not every request) to preserve cache.

pub async fn run(&mut self, user_input: String) -> Result<Vec<AgentEvent>>

Run chat loop with tool execution (streaming version).

pub fn set_messages(&mut self, messages: Vec<Message>)

Restore message history (for session continue/resume)

pub fn get_messages(&self) -> &[Message]

Get current messages (for session saving)

pub fn get_tools(&self) -> &[Arc<dyn Tool>]

Get available tools

pub fn get_system_prompt(&self) -> &str

Get system prompt

pub fn get_token_counts(&self) -> (u64, u64)

Get current token counts

pub fn clear_history(&mut self)

Clear message history

pub fn message_count(&self) -> usize

Get message count

pub async fn add_mcp_server( &mut self, name: &str, config: McpServerConfig, ) -> Result<()>

动态添加 MCP 服务器

Example

use matrixcode_core::mcp::McpServerConfig;
 
let config = McpServerConfig::stdio("npx", vec!["-y", "@playwright/mcp@latest"]);
agent.add_mcp_server("playwright", config).await?;

pub async fn remove_mcp_server(&mut self, name: &str) -> Result<()>

移除 MCP 服务器

pub async fn mcp_server_status(&self) -> Vec<ServerStatus>

获取 MCP 服务器状态列表

pub async fn start_mcp_server( &self, name: &str, ) -> Result<Vec<Arc<McpToolWrapper>>>

启动指定的 MCP 服务器

impl Agent

pub fn has_pending_inputs(&self) -> bool

Check if there are pending inputs waiting to be processed.

pub fn take_pending_inputs(&mut self) -> Vec<String>

Get and clear all pending inputs.