Comparison
goose vs Kiro
How the two compare on the things that decide whether your setup carries over: open standards, license, models and where they run.
| goose | Kiro | |
|---|---|---|
| Maker | Block, now at the Agentic AI Foundation | AWS |
| License | Apache-2.0 | Proprietary |
| Models | Many providers, including local models | See vendor docs |
| Runs in | Desktop, Terminal, API | IDE, CLI |
| MCP | MCP: supported | MCP: supported |
| Agent Skills | Agent Skills: supported | Agent Skills: supported |
| AGENTS.md | AGENTS.md: supported | ?AGENTS.md: not yet verified |
| ACP | ACP: supported | ACP: supported |
Key differences
- goose is open source (Apache-2.0); Kiro is proprietary. If you need to self-host, audit or fork the harness, that settles it.
- goose also runs in Desktop, Terminal, API; Kiro also runs in IDE, CLI.
- goose supports AGENTS.md; we have not verified it for Kiro.
- Both support MCP, Agent Skills, ACP, so MCP servers, skills and repo instructions you set up carry over if you switch.
In their own words
FAQ
- Is goose or Kiro open source?
- goose is open source under Apache-2.0. Kiro is proprietary.
- Do goose and Kiro support MCP servers?
- goose: yes. Kiro: yes.
- Can I reuse Agent Skills between goose and Kiro?
- Yes. Both read the open Agent Skills (SKILL.md) format, so the same skill folder works in both.
Set up MCP servers in either one
More comparisons
Reviewed Oct 5, 2026.