Cline

Model-agnostic coding agent across editor and terminal workflows with explicit approval and MCP control.

Type: IDE extension / CLI / SDK

Best for: Teams that want provider flexibility, explicit approvals, and configurable agent infrastructure

Official docs: https://docs.cline.bot/cline-overview

Last verified: 2026-07-13

Status: Documentation verified

Verification scope: Official Cline overview, install, config, CLI, MCP, tools, auto-approve, and troubleshooting docs were reviewed. Commands were not executed locally in this repository.

Overview

Cline is an AI coding agent that lives in the editor and terminal. The official docs describe it as able to read and write files, run terminal commands, use a browser, and operate through explicit approval. It also supports different provider paths, local models, and richer agent infrastructure such as plugins, hooks, and workflows.

Best Suited For

  • teams that want provider choice instead of a single vendor lock-in
  • developers who want explicit approval before every risky action
  • editor users who also want a real CLI and headless automation path
  • advanced teams that may want plugins, hooks, local models, or SDK-based extensions

Less Suited For

  • teams that want the simplest possible setup
  • organizations that do not want to manage provider configuration
  • environments where plugin, MCP, or auto-approve policy has not been defined

Confirmed Capabilities

  • editor and terminal workflows
  • provider authentication with the built-in service or bring-your-own-key flows
  • CLI usage with interactive and structured-output modes
  • MCP support through .cline/mcp.json
  • configurable rules, skills, hooks, plugins, agents, and workflows
  • explicit approval controls, including auto-approve and YOLO mode

Limitations

  • the feature surface is broad, which increases policy drift risk
  • provider flexibility means rollout quality depends on team conventions
  • auto-approve and YOLO mode need careful governance
  • this page is documentation-verified only, not locally exercised

Supported Environments

  • VS Code
  • Cursor
  • JetBrains
  • Windsurf
  • VSCodium
  • terminal CLI
  • SDK and Kanban according to official install docs

The official docs also document local-model usage through runtimes such as Ollama, LM Studio, and Atomic Chat.

Installation

CLI install

npm install -g cline

First interactive run

cline

Other documented install paths

The install docs also describe IDE extensions, SDK usage, and Kanban workflows. Use the official install guide when choosing between those surfaces.

Authentication

The CLI overview lists cline auth as the provider authentication entry point.

cline auth

The official docs describe both Cline-managed access paths and bring-your-own-provider credentials.

First Working Example

cline "explain this repository and identify one low-risk improvement"

For an interactive session:

cline

Expected behavior:

  • Cline starts in terminal chat or TUI mode
  • prompts before risky actions
  • keeps approval with the human unless you widen policy settings

Common Commands

The official CLI overview and reference confirm these high-value commands:

cline --help
cline auth
cline config
cline mcp
cline doctor
cline history

The TUI docs also confirm:

cline -i
cline --json "list TODO comments"

Repository Instructions

Cline supports multiple repository-level control surfaces. The official CLI reference shows project files such as:

  • .cline/rules/
  • .cline/skills/
  • .cline/hooks/
  • .cline/plugins/
  • .cline/mcp.json
  • .cline/agents.yaml

For teams, this is powerful but easy to overdo. Start with a small rules and MCP baseline before rolling out hooks or plugins broadly.

Configuration

The official config docs describe two scopes:

  • global configuration in ~/.cline/
  • project configuration in .cline/

The docs also note that global provider settings, global settings, and MCP settings are stored under ~/.cline/data/settings/.

Permission Model

The official overview explicitly says every action requires explicit approval by default. The docs also describe:

  • Auto Approve for routine operations
  • YOLO Mode for no-confirmation execution
  • command and tool controls in CLI and UI flows

This makes Cline safer by default than some agent tools, but only if you do not casually enable broad auto-approval.

MCP / Integration Support

Official docs confirm MCP support in both the general overview and the dedicated MCP docs. The docs show CLI MCP config at:

  • ~/.cline/mcp.json

MCP tools are loaded alongside built-in tools, which means each configured server directly expands what the agent can do.

Real Workflow Demonstration

Scenario

Your team wants a model-flexible coding agent but still wants human approval on risky steps.

Repository context

A local application repository opened in an editor or terminal.

Request

Explain this module, trace the main data flow, and propose one safe refactor before modifying anything.

Expected AI behavior

  • reads the code
  • proposes a plan
  • asks for approval before file writes or command execution
  • can pull in MCP tools if explicitly configured

Human review checkpoint

  • confirm the provider and model are the intended ones
  • review the proposed plan before any writes
  • verify that auto-approve is not hiding risky actions

Validation step

Run the local build or tests after the edit.

Team Adoption Guidance

  • standardize which provider path is approved
  • define whether auto-approve is allowed and for which command classes
  • keep .cline/ configuration reviewable in code review
  • allow MCP only from an approved server list

Security Considerations

  • approval is a control boundary; do not weaken it casually
  • plugins, hooks, and MCP servers all expand the execution surface
  • local-model support is useful, but teams still need policy around what data can leave the workstation
  • proxy and firewall configuration may be required in enterprise environments

Troubleshooting

  • If provider setup fails, run cline auth and cline doctor.
  • If enterprise networking blocks access, use the official proxy guidance.
  • If behavior differs across editor and CLI surfaces, inspect global and project config together.

Alternatives

  • Continue for another open and configurable CLI-plus-editor stack
  • Cursor for a more productized IDE-first experience
  • Claude Code or OpenAI Codex for more opinionated terminal-first agents

Verification Status

  • Status: Documentation verified
  • Last verified: 2026-07-13
  • Scope: official docs reviewed for install, config, CLI commands, approvals, MCP, and supported surfaces
  • Not locally tested: npm install, provider auth, editor extension behavior, and local-model setup

Sources