コンテンツにスキップ

CLI-focused Recommended Coding Agents

This page collects coding agents selected specifically for terminal-first (CLI) use and focuses on agents that are practical to configure, run, and extend entirely from the command line.

Quick summary

  • Best overall CLI flexibility: OpenCode
  • Best for official/commercial ecosystems on CLI: Codex CLI
  • Best lightweight daily editing tool: Aider
  • Best terminal UX/TUI experience: Crush
  • Best for MCP/Extensions-driven workflows: Goose / Qwen Code

Top recommendations (CLI perspective)

  1. OpenCode — Broad multi-provider support and rich extension surface. Easy to integrate local models and custom providers; strong CLI primitives for sessions and webfetch.
  2. Goose — Strong MCP/extension ecosystem, skills marketplace, session persistence, and automation-oriented workflow design.
  3. Codex CLI — OpenAI's official CLI with plugin/marketplace momentum and Ollama/local integrations suitable for commercial deployments.
  4. Qwen Code — Fast-moving multi-provider agent with MCP, extensions, skills, WebSearch/WebFetch, and checkpointing.
  5. Crush — Polished TUI and multi-model/LSP support for users who spend most of their time in a terminal.
  6. Plandex — Strong long-running planning, branching, and context-management workflow for larger implementation plans.
  7. Aider — Lightweight, low-friction CLI-first assistant for everyday edits and pair-programming in the terminal.

Evaluation method

The comparison scope is intentionally narrow: publicly available coding agents that can be used from the CLI and can switch models or providers at the API/configuration level.

Scores prioritize agent-platform capabilities, not only editing quality. Tools with MCP, plugins, skills, marketplace/distribution, session management, auto-compaction, web search/fetch, and flexible provider support score higher than editor-only tools.

Feature cells use C/M/S = Completeness / Maturity / Stability, each on a 0-5 scale.

  • Completeness: how fully the feature exists as a user-facing product capability.
  • Maturity: documentation quality, release history, and implementation age.
  • Stability: recent issue pattern, known rough edges, and operational predictability.

Overall score

Rank Tool Score Subscription bonus Best fit
1 OpenCode 88 +5 Best overall CLI agent platform with broad provider, MCP, skills, plugin, web, and session coverage
2 Goose 86 +5 Best for MCP/extension-driven automation workflows
3 Codex CLI 84 +5 Best official/commercial ecosystem choice for OpenAI-oriented teams
4 Qwen Code 79 +4 Fast-moving multi-provider agent with strong extension potential
5 Crush 77 +5 Best terminal UX/TUI choice with multi-model and LSP support
6 Plandex 74 +0 Strongest long-running planning and context-management workflow
7 Aider 71 +3 Most practical lightweight terminal editor assistant 
xychart-beta
    title "Overall CLI Coding Agent Score"
    x-axis [OpenCode, Goose, Codex, QwenCode, Crush, Plandex, Aider]
    y-axis "score" 0 --> 100
    bar [88, 86, 84, 79, 77, 74, 71]

Feature comparison

Tool MCP Plugin Skills Marketplace Resume Sessions
OpenCode 5/4/4 5/4/4 5/4/4 1/1/3 5/4/3 5/4/3
Goose 5/4/4 5/4/4 5/4/4 5/4/4 5/4/4 5/4/4
Codex CLI 5/4/4 5/4/3 5/4/3 5/4/3 5/4/3 5/4/3
Qwen Code 5/3/3 5/3/3 5/3/4 4/3/3 4/3/3 2/2/3
Crush 5/4/3 0/0/0 4/4/4 0/0/0 4/4/3 4/4/3
Plandex 0/0/0 0/0/0 0/0/0 0/0/0 4/5/4 5/5/4
Aider 0/0/0 0/0/0 0/0/0 0/0/0 4/5/4 2/4/4
Tool Auto-compaction Tools WebSearch WebFetch Read Edit Subscription usable
OpenCode 5/4/4 5/4/4 5/4/4 5/4/4 5/4/4 5/4/4 Yes
Goose 5/4/4 5/4/4 4/4/3 4/4/4 5/4/4 5/4/4 Yes
Codex CLI 5/4/3 5/4/3 4/4/3 3/3/3 5/4/3 5/4/3 Yes
Qwen Code 2/2/3 5/3/3 5/3/3 4/3/3 5/3/3 5/3/3 Yes
Crush 4/3/3 4/4/3 1/1/2 0/0/0 4/4/4 4/4/4 Yes
Plandex 5/5/4 3/4/4 0/0/0 3/4/4 5/5/4 4/5/4 Unconfirmed
Aider 5/5/4 4/5/4 0/0/0 4/5/4 5/5/4 5/5/4 Yes

Per-tool CLI notes

OpenCode

  • CLI-first extensibility: supports 75+ providers, plugins/skills, native websearch/webfetch.
  • Full session features (continue/fork/share) and auto-compaction policies.
  • Best when you want one terminal-first agent platform that can cover provider switching, MCP, skills, web tools, and long-running sessions.
  • Caveat: marketplace-style distribution is less mature than Goose or Codex CLI, and compaction/model discovery/TUI issues still need operational monitoring.

Goose

  • Strong MCP-centered architecture with extension directory, skills marketplace, session persistence, memory-oriented features, and auto-compaction.
  • Best when you want to build repeatable automation workflows rather than only edit code interactively.
  • Caveat: some provider bridge modes may not expose the full Goose extension ecosystem, and the project is still changing quickly.

Codex CLI

  • Official OpenAI CLI. config.toml and Ollama integration enable custom provider/local model workflows.
  • Growing plugin/marketplace ecosystem and favorable for commercial/enterprise use.
  • Best when your team already uses ChatGPT/Codex subscriptions or needs an official commercial ecosystem.
  • Caveat: custom provider and local-model switching still have rough edges, especially around history visibility, model switchers, and compact behavior.

Qwen Code

  • Fast-growing multi-provider CLI with modelProviders, MCP, extensions, skills, WebSearch/WebFetch, checkpointing, and read/edit tools.
  • Best when you want a broad modern agent surface and can tolerate a younger implementation.
  • Caveat: recent issues around auth, provider display, and MCP/provider connections mean environment-level validation is important.

Aider

  • Practical CLI modes (/ask, /architect, /web, --read) and strong Git/diff editing flow compatibility.
  • Low adoption cost and predictable behavior for everyday edits.
  • Best when you mainly want fast, predictable terminal-based edits rather than a full agent platform.
  • Caveat: limited MCP/plugins/skills/marketplace extensibility compared to OpenCode, Goose, Codex CLI, or Qwen Code.

Crush

  • Polished TUI from Charmbracelet, with multi-model and LSP support for terminal-first power users.
  • Agent skills, LSP integration, and session-based context make it a strong terminal UX option.
  • Caveat: plugin, marketplace, and native websearch/webfetch capabilities are limited compared to the platform-style agents.

Plandex

  • Strong long-running planning model with plans, branches, model packs, role-based models, and context-window management.
  • Best when the work is a long multi-step implementation plan rather than ad-hoc pair editing.
  • Caveat: weak MCP/plugin/skills/marketplace/native websearch coverage under this evaluation model.

How to choose (CLI-focused)

  • Maximize freedom and extensibility: OpenCode
  • Leverage existing OpenAI subscriptions and enterprise readiness: Codex CLI
  • Minimal friction daily CLI edits: Aider
  • Terminal UX/TUI priority: Crush
  • Build extension-driven workflows (MCP/Skills): Goose / Qwen Code

Validation checklist (before adoption)

  1. Test Web Tools / compaction / session behavior when swapping providers
  2. Validate real-world switching for local models (Ollama etc.) and profile management
  3. Confirm plugin/skill installation steps and permission model via CLI
  4. Verify session persistence, resume/fork behavior and storage requirements