Claude Code Alternatives: Terminal and IDE Agents Compared

The right Claude Code alternative depends on why you are leaving, or adding a second agent. If you want another vendor’s agent in the same terminal, look at Codex CLI or Gemini CLI. If you want open source and a free choice of model, look at OpenCode, Aider, Goose, Qwen Code or Cline. If you would rather the agent live inside your editor, Cursor, Junie and Cline are built for that. If you want to hand over a whole ticket and review a pull request later, Devin is the delegation option. Whichever you choose, three things carry over almost untouched: an AGENTS.md file, your MCP servers, and a task list kept outside any one agent’s chat.

This page is a map by fit, one short paragraph per tool, each linking to a longer head-to-head where one exists. For the full side-by-side of every agent against six questions, read the best AI coding agents. No prices here: plans change often, so check each vendor’s current page.

Why people look for an alternative

• Model choice: Claude Code runs Claude models. Several alternatives run models from any provider, or a local model on your own machine.
• Open source: some teams need to read, audit or change the agent itself, or run it where proprietary binaries are not allowed.
• Where you work: a terminal agent visits your tools; an IDE agent shows each diff inside the file you are reading.
• How much you delegate: pairing in a session and handing off a ticket to a cloud agent are different jobs.
• Account and sign-in: the agent has to fit how your company already pays for AI, whether that is a ChatGPT plan, a Google Cloud project or an API key.

Another vendor’s terminal agent

Codex CLI (OpenAI)

The Codex repository↗ (opens in a new tab) describes Codex CLI as a coding agent from OpenAI that runs locally on your computer, licensed Apache-2.0. You install it with a script, npm or Homebrew, then sign in with a ChatGPT account on a paid plan or use an API key. It reads AGENTS.md, contains commands in a sandbox, and the same agent also runs in VS Code, a desktop app and the cloud. Suits people who already pay for ChatGPT and want an AGENTS.md-first agent. Compared in Claude Code vs Codex.

Gemini CLI (Google)

Google’s open-source terminal agent reads GEMINI.md, has a plan mode, sandboxing and MCP support. Check how you will sign in before you switch: Google’s Code Assist deprecation notice↗ (opens in a new tab) says that from June 18, 2026, Login with Google no longer works for Gemini CLI on the Code Assist for individuals, Google AI Pro and Google AI Ultra tiers, and points those users to Antigravity. A Gemini API key, Vertex AI, or a company’s Code Assist Standard or Enterprise license still work. Compared in Gemini CLI vs Claude Code vs Codex CLI, with Google’s newer tool in Antigravity vs Claude Code.

Open source, any model

OpenCode

An MIT-licensed terminal agent with an agent loop much like Claude Code’s, Build and Plan agents, and models from many providers or a local server. It suits people who like how Claude Code works but want to pick the model. See what OpenCode is and OpenCode vs Claude Code.

Aider

An Apache 2.0 pair programmer that edits the files you add and commits every change to Git with a descriptive message, so /undo is a Git revert. It works with almost any model, and its documentation does not describe MCP support. Suits people who want each AI edit as its own commit. See Aider vs Claude Code.

Goose

An open-source agent with a desktop app, a CLI and an API, now under the Agentic AI Foundation. Every extension is an MCP server, recipes package a workflow as a file, and it runs on nearly any provider. Note that it starts in Completely Autonomous mode. See Goose, the open-source agent.

Qwen Code

The Qwen team’s agent for the terminal, a desktop app, an experimental web UI and editor plugins, licensed Apache 2.0. Its repository↗ (opens in a new tab) says it was originally based on Gemini CLI v0.8.2 and has developed independently since v0.1, and that it speaks the OpenAI, Anthropic, Gemini and Qwen APIs, including local models through Ollama or vLLM. It reads QWEN.md for project instructions and supports MCP, subagents and a headless qwen -p mode. The free Qwen OAuth sign-in was discontinued on April 15, 2026, so plan on an API key or an Alibaba Cloud plan.

OpenHands

An MIT-licensed agent platform with a browser control center, a CLI, a Python SDK and a hosted cloud, which can also drive Claude Code, Codex or Gemini CLI as the agent. It suits teams that want to run agents in Docker sandboxes on their own machines. See OpenHands explained.

An agent inside your IDE

Cursor

An AI code editor built around its agent, with Agent, Plan and Ask modes, checkpoints, rules in .cursor/rules, and cloud agents that run in their own virtual machines. It reads AGENTS.md, and its Cursor CLI↗ (opens in a new tab) brings the same agent to the terminal. See Claude Code vs Cursor.

Cline

An open-source agent that started as a VS Code extension and now also ships as a CLI with a terminal UI, an SDK, a Kanban view for parallel agents in Git worktrees, and a desktop app. You bring your own model, including an Anthropic key or a Claude subscription. See Cline, Roo Code and Kilo Code vs Claude Code.

Junie (JetBrains)

JetBrains’ agent runs inside AI Chat, as an IDE plugin and as a CLI that borrows the IDE’s inspections and refactorings when both are open on the same project. It reads .junie/AGENTS.md first. Suits teams who live in IntelliJ, PyCharm or Rider. See Junie vs Claude Code.

An agent you delegate to

Devin (Cognition)

Devin Cloud works on its own machine and takes tasks from Slack, Linear, Jira or the web app, then returns a pull request. The same brand now covers Devin Desktop, the IDE formerly called Windsurf, a Devin CLI and Devin Review. Suits teams with a queue of well-scoped tickets. See Devin vs Claude Code and Windsurf vs Claude Code.

What to keep when you switch

Moving agents is cheap if your instructions and connections live in files, not in one tool’s settings. Three things travel well.

• AGENTS.md: the AGENTS.md format↗ (opens in a new tab) is a plain Markdown file for build commands, conventions and no-go areas, and most agents on this page read it. Claude Code reads it too, and a CLAUDE.md that contains @AGENTS.md keeps one shared file for every tool. Gemini CLI defaults to GEMINI.md, so point it at the shared file; Qwen Code reads QWEN.md and also picks up an existing AGENTS.md. Which tool reads what is in AI context files compared.
• Rules files: .cursor/rules, .junie/AGENTS.md and Cline’s rules are tool-specific. Keep the substance in AGENTS.md and let each rules file hold only what is unique to that tool.
• MCP servers: the server URL and its sign-in stay the same; only the config file changes. Each agent stores servers in its own place, so you add them again, but nothing on the server side moves.
• Skills: several agents load SKILL.md folders in the Agent Skills format, so a skill written for Claude Code often works elsewhere. See agent skills.

What does not travel: permission modes, hooks, checkpoints and session history. Each agent has its own model for approvals and undo, so relearn it before you give a new agent write access to anything important.

# Claude Code
claude mcp add --transport http fenbs https://fenbs.ai/api/mcp

# Codex CLI
codex mcp add fenbs --url https://fenbs.ai/api/mcp

# Gemini CLI
gemini mcp add --scope user --transport http fenbs https://fenbs.ai/api/mcp

Keep the work list outside the agent

Session history is the one thing you lose on every switch, and it is where the plan usually lives. fenbs keeps it on a board instead. Each task has a note saying what and why, a plan, and a test status, in lanes To Do, Next Up, In Progress and Completed. Any agent above that speaks MCP connects to https://fenbs.ai/api/mcp, signing in with OAuth or a token you issue under Settings with a name, scopes and an optional expiry. History records each change under the assistant’s name, so when you move from Claude Code to Codex, or run both, the next agent reads the same tasks and the same rules. Clients that only speak stdio can use the bridge, npx -y fenbs-mcp. fenbs is deliberately small: no sprints, due dates or epics.