What is the difference between goose and Claude Code?
goose is an open source, model-agnostic AI agent with a desktop app, a CLI and an API, while Claude Code is Anthropic's agentic coding tool built for Claude models across the terminal, IDEs, a desktop app and the web. Pick goose if you want to choose your own model and own the code; pick Claude Code if you want Anthropic's tightly integrated tooling around Claude.
Both read your project, edit files, run commands, connect to MCP servers and load Agent Skills. The differences are in licensing, how you pay, how much they do without asking, and how far each goes beyond the terminal. Everything below comes from the official docs as read on 26 October 2026. For a full tour of goose on its own, see the goose guide.
How does goose work?
The goose repository describes it as "an open source, extensible AI agent that goes beyond code suggestions - install, execute, edit, and test with any LLM." It is licensed under Apache 2.0, and the README says goose "is part of the Agentic AI Foundation (AAIF) at the Linux Foundation."
You install the CLI with a script or Homebrew, pick a provider, then start a session:
curl -fsSL https://github.com/aaif-goose/goose/releases/download/stable/download_cli.sh | bash
# or: brew install block-goose-cli
goose configure # choose Configure Providers and paste a key
cd my-project
goose session
The README claims "15+ providers," naming Anthropic, OpenAI, Google, Ollama, OpenRouter, Azure and Bedrock, and says goose can "Connect to 70+ extensions via the Model Context Protocol." Extensions are how goose gets tools, and the built-in Developer extension supplies file and shell access.
goose also packages work for reuse. Recipes are YAML or JSON files with instructions, parameters and extensions, run with goose run --recipe, and goose schedule add runs them on a cron schedule. The CLI commands page also lists goose review, which reviews your current git diff.
How does Claude Code work?
Claude Code is "an agentic coding tool that reads your codebase, edits files, runs commands, and integrates with your development tools." It runs in the terminal, VS Code, JetBrains IDEs, a desktop app and the browser, and the docs say each surface "connects to the same underlying Claude Code engine."
curl -fsSL https://claude.ai/install.sh | bash
# or: brew install --cask claude-code
cd your-project
claude
Around the core agent you get CLAUDE.md instructions, skills, subagents, hooks, MCP servers and plugins, plus GitHub Actions and GitLab CI/CD integrations. For scripts, claude -p "query" runs one task and exits.
Claude Code runs Claude models, through a Claude subscription, an Anthropic Console account, or a cloud provider such as Amazon Bedrock, Google Cloud's Agent Platform or Microsoft Foundry. It is not open source: the license file in Anthropic's claude-code repository reads "All rights reserved" and points to Anthropic's Commercial Terms of Service.
How do goose and Claude Code compare side by side?
| goose | Claude Code | |
|---|---|---|
| Maintainer | AAIF at the Linux Foundation (founded by Block) | Anthropic |
| License | Apache 2.0 | Proprietary, all rights reserved |
| Interfaces | Desktop app, CLI, API | Terminal, VS Code, JetBrains, desktop app, web |
| Models | 15+ providers, local models, subscriptions through ACP providers | Claude models via Anthropic or supported cloud providers |
| Cost of the tool | Nothing; you pay your model provider | Paid Claude plan, Console account or cloud provider |
| Default autonomy | Completely Autonomous | Manual: asks before commands and edits |
| Project instructions | AGENTS.md and .goosehints |
CLAUDE.md, or AGENTS.md when there is no CLAUDE.md |
| Agent Skills | .agents/skills/, also reads .claude/skills/ |
.claude/skills/, ~/.claude/skills/ |
| Reusable workflows | Recipes, scheduled with goose schedule add |
Skills, plugins, hooks, scheduled tasks |
| Scripting | goose run -t "prompt" |
claude -p "prompt" |
How do pricing and model choice differ?
This is the main reason people compare the two.
goose costs nothing as software. Your bill depends on the model you connect: an API key from Anthropic or OpenAI, a gateway such as OpenRouter, or a local model through Ollama with no API costs. The ACP providers page adds another route: it says ACP providers let you use goose with an existing Claude Code or ChatGPT Plus or Pro subscription, with "no per-token API costs." For Claude, that means installing the claude-agent-acp adapter and signing in with the claude CLI first. The docs note that goose session resume and goose session fork do not work with ACP providers yet.
Claude Code is included in paid Claude plans. As of October 2026, Claude's pricing page says "Claude Code is included in all paid plans," marks it as not included on Free, and lists Pro at $20 a month billed monthly ($17 a month billed annually) and Max from $100 a month. You can also pay per token through a Console account. The Claude Code pricing guide covers the details.
If you want to switch between OpenAI, Google and local models from one agent, goose is built for that. If you already pay for Claude and mostly use Claude models, Claude Code gives you Anthropic's own harness for them.
How do permissions and safety defaults compare?
The defaults point in opposite directions, so check them before you start.
goose has four permission modes: Completely Autonomous (/mode auto), Smart Approval (/mode smart_approve), Manual Approval (/mode approve) and Chat Only (/mode chat). The docs state: "Autonomous Mode is applied by default." On a repo you care about, switch to Smart Approval or Manual Approval first.
Claude Code's permission modes are default, acceptEdits, plan, auto, dontAsk and bypassPermissions. The default mode, labeled Manual in the CLI, "Prompts for permission on first use of each tool." Claude Code also has an optional OS-level sandbox for shell commands, using Seatbelt on macOS and bubblewrap on Linux and WSL2. The docs say "The sandbox is off by default," and you turn it on with /sandbox. See the permissions guide and sandbox guide.
How do instructions, skills and MCP compare?
Instructions. goose's hints docs say it "looks for AGENTS.md then .goosehints files by default," and the CONTEXT_FILE_NAMES environment variable changes the list. Claude Code's memory docs say it reads AGENTS.md "only when you have no CLAUDE.md in your working directory or above it." So one root AGENTS.md serves both. If you keep a CLAUDE.md, tell goose to read it:
export CONTEXT_FILE_NAMES='["AGENTS.md", "CLAUDE.md", ".goosehints"]'
The CLAUDE.md vs AGENTS.md guide covers the other direction, importing AGENTS.md from CLAUDE.md.
Skills. Claude Code's skills docs say its skills "follow the Agent Skills open standard," and load from .claude/skills/ and ~/.claude/skills/. goose's primary folders are .agents/skills/ and ~/.agents/skills/, and its skills page lists .claude/skills/ and ~/.claude/skills/ as backward-compatible locations. A skill written for Claude Code works in goose without moving. See the Claude Code skills guide.
MCP. Both use MCP for tools. goose calls MCP servers extensions and adds them through goose configure or the desktop app. Claude Code adds them with claude mcp add or a project .mcp.json, covered in the Claude Code MCP guide. The config formats differ, so you set each up separately.
Which should you pick?
Neither is right for everyone. Start with how you want to pay and how much you want to tune.
goose tends to fit when:
- You want an Apache 2.0 agent under a foundation that you can audit or fork.
- You want to switch between providers, or run local models with no API bill.
- You like a desktop app as well as a CLI.
- You want recipes you can share and schedule.
Claude Code tends to fit when:
- You already pay for a Claude plan, or want Anthropic's harness for Claude models.
- You want one engine across terminal, VS Code, JetBrains, desktop and web.
- You want permission prompts on by default, plus an optional OS-level sandbox.
- You want built-in GitHub Actions and GitLab CI/CD integrations, hooks and plugins.
They also work together. goose can drive Claude Code's subscription through an ACP provider, and both read AGENTS.md and .claude/skills/, so trying the other one costs little. Other open source options are covered in OpenCode vs Claude Code.
Keep your prompts and skills portable
Your skills and house rules matter more than the agent that reads them. Kipt keeps prompts and Agent Skills as Markdown in a private GitHub repo you own, with {{name}} blanks and Cmd+K search, and installs every skill into Claude Code or Codex with one command: npx skills add https://github.com/<you>/<repo>.git --all -g. To turn a prompt into a skill folder first, try the free SKILL.md generator.