# Celeste CLI MCP server

> An agentic terminal AI assistant that performs code reviews, semantic searches, and autonomous development tasks using a rich TUI.

- Repository: https://github.com/whykusanagi/celeste-cli
- Page: https://hostedmcp.org/servers/whykusanagi/celeste-cli
- Transport: —
- Stars: 1
- License: MIT

## Environment variables

| Name | Required | Secret | Description |
|---|---|---|---|
| `CELESTE_API_ENDPOINT` | no | no |  |
| `CELESTE_API_KEY` | no | no |  |
| `GROK_API_KEY` | no | no |  |
| `OPENAI_API_KEY` | no | no |  |
| `TAROT_AUTH_TOKEN` | no | no |  |
| `VENICE_API_KEY` | no | no |  |

## Tools (10)

- `goal`
- `type`
- `workspace`
- `task_id`
- `depends_on`
- `max_turns`
- `isolate_worktree`
- `background_after`
- `persona`
- `message`

## FAQ

### Does Celeste support different LLM providers?

Yes, it supports multiple providers including Sakana AI, Grok/xAI, OpenAI, Anthropic, Gemini, Venice.ai, Vertex AI, OpenRouter, and local OpenAI-compatible servers.

### How does Celeste handle code analysis?

It uses a code graph with semantic search via MinHash and BM25, and tree-sitter parsing for accurate call-graph edges in supported languages.

### Can I run Celeste in a sandbox?

Yes, you can opt-in to a sandbox where bash runs under the OS sandbox (seatbelt on macOS, bubblewrap on Linux) to restrict file writing and network access.
