LevitasMCP

Rust Model Context Protocol License: GPL v3 Platform

Levitas-MCP (from Latin levitas for "lightness") is a premium, high-performance Model Context Protocol (MCP) server written in Rust. It acts as a lightweight, zero-friction bridge that elevates simple LLMs into fully autonomous AI coding agents by supplying advanced filesystem indexers, semantic memory, safe git access, and hierarchical goal-planning graphs.

---

✨ Features

• 🧠 Semantic Memory (SQLite + Vectors):
  • Powered by a local SQLite store (ultra_memory.db) with Ollama vector embedding fallback.
  • Dynamically remembers context (e.g. "that Rust project about Discord bots") instead of relying on exact string matching.
  • Automatically summarizes old memories to optimize context size.
• 🌳 Codebase Dependency Analysis (show_dependencies):
  • Recursively maps file imports and module structure.
  • Out-of-the-box support for Rust (mod files), Python (imports/packages), and JS/TS (ES6 imports & CommonJS require).
  • Outputs beautiful, structured ASCII trees.
• 🔍 Boundary-Constrained Usages (find_references):
  • Scans the workspace to locate occurrences of any function, struct, or variable.
  • Excludes definition and declaration lines, returning only actual reference usages with line numbers.
• 📈 Hierarchical Goal Planning:
  • Organizes agent workloads in a persistent hierarchical tree (goal_add, goal_update, goal_list).
  • Features cascading deletions to prune obsolete subtask branches.
  • Visual progress tracking using [ ] Pending, [/] In Progress, and [x] Completed indicators.
• 📂 structural workspace indexing:
  • Instantly scans and indexes traits, classes, functions, and imports in files without reading them individually.
• 🚀 Native Git Tools:
  • Exposes safe, built-in commands (git_status, git_diff, git_commit, git_branch, git_log) directly to the agent, bypassing arbitrary shell execution risks.
• 🔒 Sandbox & Resource Limits:
  • Restricts background scheduler processes using safe CPU limits, memory limits, task timeouts, and setrlimit hooks.

---

🛠️ Supported Clients

Levitas-MCP includes a dynamic installer supporting 20 distinct client environments across 67 configurations on macOS, Windows, and Linux.

Client Type	Target Clients	Flags
Editors & IDEs	Cursor, Zed, Windsurf, PearAI	--cursor, --zed, --windsurf, --pearai
IDE Extensions	Cline, Roo Code, Continue, Sourcegraph Cody	--cline, --roo, --continue, --cody
Desktop App Agents	Claude Desktop, Jan AI, AnythingLLM	--claude, --jan, --anythingllm
CLI Agents (Modern)	Claude Code, Qwen Code, GitHub Copilot CLI, Aider	--claudecode, --qwencode, --copilotcli, --aider
Google Ecosystem	Antigravity (Agy CLI & Desktop IDE)	--antigravity
Sourcegraph / OpenAI	Amp Code CLI, Codex CLI	--amp, --codex
Generic / Others	mcp.json / mcp-servers.json (e.g. for fastmcp, lobe-cli, etc.)	--generic

---

🚀 Quick Start

1. Build and Install

Clone the repository and run the installer in your terminal. It will automatically build the Rust workspace in --release mode and register the server for all compatible discovered clients:

On macOS and Linux:

chmod +x install.sh
./install.sh

On Windows:

install.bat

2. Targeting Specific Clients

If you want to skip auto-detection and install only for specific agents:

./install.sh --claudecode --aider --cursor

To configure all 20 supported clients simultaneously:

./install.sh --all

---

⚙️ Environment Variables

The installer automatically captures and forwards current environment credentials to the client configuration. Make sure these are set in your environment before running the installation:

• OPENAI_API_KEY: API token for OpenAI models and default fallback.
• MCT_EMBEDDINGS_API_KEY: Custom endpoint API token for vector memory.
• OLLAMA_API_BASE: Endpoint for Ollama vector fallback (Defaults to http://localhost:11434).
• EMBEDDING_API_URL / EMBEDDING_MODEL: Specific models and endpoints for embeddings.

---

📄 License

Levitas-MCP is licensed under the GNU General Public License v3.0. See the LICENSE file for more details.