Agents
Cursor Agent Overview: Tools, Models & How It Works
Cursor Agent is an assistant that completes coding tasks in your repo: it follows instructions (system prompt plus rules), uses tools (search, read/edit files, terminal, browser, web) and runs on the model you pick. Open it from the sidepane with Cmd+I (Ctrl+I on Windows). There is no cap on tool calls per task; Cursor tunes prompts and tools per model.
On this page
What three pieces make up an agent run?
- Instructions
- System prompt plus your
.cursor/rulesand fetched rule files. - Tools
- Search, read/edit files, shell, browser, web, images and clarifying questions.
- Model
- The model you select; Cursor adapts tooling per model.
Which tools can Agent use on my codebase?
- Tool
- Semantic search
- What it does
- Find code by meaning across your indexed repo.
- Tool
- File search
- What it does
- Locate files and exact patterns by name or path.
- Tool
- Read / edit files
- What it does
- Load content (including images for vision models) and apply patches.
- Tool
- Shell
- What it does
- Run terminal commands using your default terminal profile.
- Tool
- Browser
- What it does
- Navigate, click and screenshot running apps for verification.
- Tool
- Web
- What it does
- Search and fetch external documentation.
- Tool
- Ask questions
- What it does
- Pause for your answer while continuing background work.
| Tool | What it does |
|---|---|
| Semantic search | Find code by meaning across your indexed repo. |
| File search | Locate files and exact patterns by name or path. |
| Read / edit files | Load content (including images for vision models) and apply patches. |
| Shell | Run terminal commands using your default terminal profile. |
| Browser | Navigate, click and screenshot running apps for verification. |
| Web | Search and fetch external documentation. |
| Ask questions | Pause for your answer while continuing background work. |
How does Agent relate to other modes?
Agent is the doer: it edits and runs commands. Ask stays read-only for explanation. Plan Mode, Debug Mode and Design Mode are specialized agent workflows for planning, runtime-evidence debugging and visual UI iteration. Agent Review checks diffs after edits. Pick the mode that matches whether you are exploring, executing, debugging or reviewing.
What habits make Agent runs reliable?
- @-mention the files and folders that matter instead of making Agent search blind.
- State success criteria: a test command, URL or expected behavior.
- Approve terminal commands that touch production, secrets or billing.
- Start a fresh chat when context fills with failed attempts.
Frequently asked questions
Is there a limit on how many tool calls Agent can make?
No. Agent can iterate with as many tool calls as the task requires, subject to your plan usage and model limits.
Does Agent use my whole repo as context every time?
It retrieves relevant chunks via search and @-mentions rather than loading the entire repository into context.
Where do I open Agent in the UI?
Use Cmd+I (Ctrl+I) for the sidepane agent, or open the dedicated Agents Window for multi-agent workflows.
Sources & last verified
Cursor ships frequently. Facts verified against primary sources on June 24, 2026.