Celeste CLI is an agentic terminal AI assistant developed by Whykusanagi for the mcp platform. This free MCP server tool is built with Go and connects to multi-provider Large Language Models, providing graph-based code intelligence to AI clients. It integrates with compatible AI clients like Claude Code to offer specialized code analysis capabilities. Celeste CLI manages its own configuration, including API keys and model settings, ensuring independent operation. It targets developers and AI users seeking advanced code understanding and quality assurance through structural analysis and semantic searching, enhancing various code-related workflows effectively.

What does Celeste CLI connect to?

Celeste CLI functions as an MCP server, designed to connect with multi-provider Large Language Models. Its confirmed compatible AI client is Claude Code, establishing a direct link for AI-driven code analysis. While general MCP server specifications support STDIO, SSE, and Streamable HTTP transports, the integration with Claude Code primarily utilizes STDIO transport. This setup allows Celeste CLI to serve its graph-based code intelligence capabilities directly to the connected AI client, facilitating advanced code understanding tasks within the terminal environment.

Core features and tools

Celeste CLI exposes several capabilities as callable tools when integrated with Claude Code. These include Graph Code Review, which performs structural analysis to detect issues like stubs, lazy redirects, error swallowing, and hardcoded values. Semantic Code Search finds functions by concept using MinHash similarity. Dependency Analysis maps package connectivity and identifies isolated code. Project Memory persists learned context across sessions in a ".grimoire" store, and Context auto-detects project configuration with staleness tracking and Git-stamped metadata. Specific skills are "/celeste-review", "code_search", and "save_memory".

Who is it for?

Celeste CLI is designed for developers and AI users who engage in code-related tasks and utilize AI assistants. Its capabilities support those needing structural code analysis, semantic code searching, and dependency mapping. It benefits individuals looking to enhance code quality, navigate large codebases by concept, and maintain persistent project understanding across development sessions. Skip this if your primary need is not focused on AI-assisted graph-based code intelligence or if you do not work with compatible AI clients like Claude Code.

How to get started

The server authenticates using an API key configured via a command. This API key is stored independently in "~/.celeste/config.json", and by default, Celeste CLI uses xAI/Grok.

Use cases

• Graph-Based Code Review: Developers can perform structural analysis to identify various code quality issues, including stubs, error swallowing, and hardcoded values.
• Semantic Code Search: Users can find code elements based on their conceptual meaning, enhancing code navigation and understanding within a project.
• Dependency Analysis and Project Memory: It assists in mapping code dependencies, finding isolated code, and maintaining persistent project context across development sessions through a ".grimoire" store.

Celeste CLI by Whykusanagi offers a specialized terminal AI assistant for code analysis on the mcp platform. It provides valuable tools for structural code review, semantic search, and dependency mapping, enhancing developer workflows with persistent project context. Operating as an MCP server with Claude Code, it empowers AI clients with graph-based code intelligence.