Qoder CN Agent mode features autonomous decision-making, environment awareness, and tool use. Based on the developer's coding requirements, it uses tools such as project search, file editing, and terminal to complete coding tasks end-to-end. Developers can also configure MCP tools to broaden the AI coding assistant's capability and better fit their workflows.
Agent mode has these core capabilities:
The Agent identifies intent autonomously and generates a plan for complex tasks. You can also actively trigger plan generation with /plan.
The generated plan is shown to you for review. Once confirmed, the Agent executes according to the plan.
Based on your input, the Agent generates a to-do list. If you want to add steps, simply add to your requirement, and the corresponding to-dos are added to the plan. At the bottom of the chat window, you can view each task's progress:
Qoder CN provides more than ten built-in coding tools that the Agent decides how to use autonomously, including file search, file read, directory reading, semantic symbol retrieval within the project, file modification, error retrieval, and terminal execution. While using tools, the Agent makes decisions and executes without developer confirmation or intervention, and it decides the next step based on the returned results.
Agent mode autonomously decides and executes terminal commands based on the current requirement. To ensure execution certainty, by default each command requires developer confirmation:
You can also configure an allowlist of auto-executed commands in plugin settings; configured commands run automatically without confirmation. Separate multiple commands with commas.
The exact path: on the plugin Chat settings page, in the Auto-Run section, add commands allowed to run automatically to the input field under Terminal in Agent Mode.
When MCP tools are configured, the Agent autonomously decides whether to invoke them. Each execution is confirmed first; click the execute button and the Agent calls the MCP tool and uses the returned results as context for follow-up.
For a detailed guide on using MCP in Qoder CN, see MCP.
Qoder CN supports automatically optimizing the prompt in the input box, combining the added context, conversation history, and other relevant information to further improve results.
- Project-level changes: Based on your task description, the Agent autonomously decomposes tasks and modifies multiple code files in a project. You can iterate step-by-step through multi-turn conversations or roll back to snapshots, completing the coding task collaboratively with Qoder CN.
- Automatic project awareness: Based on your task description, the Agent automatically senses information such as the project framework, tech stack, needed files, and error messages, so you don't need to add project context manually.
- Tool use: Autonomously uses over ten built-in coding tools, such as read/write file, code lookup, and error diagnosis. It also automatically discovers and uses MCP tools.
- Terminal command execution: During a coding task, the Agent autonomously decides which commands to run, writes them, and runs them in the terminal, greatly improving task execution efficiency.
Planning
The Agent identifies intent autonomously and generates a plan for complex tasks. You can also actively trigger plan generation with /plan.
The generated plan is shown to you for review. Once confirmed, the Agent executes according to the plan.
To-dos
Based on your input, the Agent generates a to-do list. If you want to add steps, simply add to your requirement, and the corresponding to-dos are added to the plan. At the bottom of the chat window, you can view each task's progress:
- Hollow circle: task not yet started.
- Spinning circle: task in progress.
- Checkbox: task completed.
Use programming tools
Qoder CN provides more than ten built-in coding tools that the Agent decides how to use autonomously, including file search, file read, directory reading, semantic symbol retrieval within the project, file modification, error retrieval, and terminal execution. While using tools, the Agent makes decisions and executes without developer confirmation or intervention, and it decides the next step based on the returned results.
Execute terminal commands
Agent mode autonomously decides and executes terminal commands based on the current requirement. To ensure execution certainty, by default each command requires developer confirmation:
- Click Run and the Agent sends the command to the IDE terminal for execution;
- Click Cancel to skip this execution, and the Agent continues planning based on your feedback.
When running terminal commands, exceptions may occur. See Terminal execution exception notes for solutions.
Call MCP tools
When MCP tools are configured, the Agent autonomously decides whether to invoke them. Each execution is confirmed first; click the execute button and the Agent calls the MCP tool and uses the returned results as context for follow-up.
For a detailed guide on using MCP in Qoder CN, see MCP.
One-click prompt optimization
Qoder CN supports automatically optimizing the prompt in the input box, combining the added context, conversation history, and other relevant information to further improve results.
- Enter your prompt: type your initial idea in the input box (for example, "Add a ticket escalation feature").
- Click the "Optimize input" button ✨: based on your input and richer context, Qoder CN generates a detailed and actionable prompt.
- Review and edit: the optimized prompt contains a clear goal, constraints, and implementation guidance. You can further refine it as needed.
- Undo or submit: revert the enhancement if needed, or submit the optimized prompt to generate a higher-quality response.