msty-admin
AI-powered administration for Msty Studio Desktop. 24 MCP tools for database insights, config sync, local model orchestration, and Claude handoff workflows.
claude mcp add --transport stdio m-pineapple-msty-admin-mcp python -m msty_admin_mcp \ --env MSTY_HOST="Msty installation host (default: 127.0.0.1)" \ --env MSTY_AI_PORT="Local AI (Ollama) port (default: 11964)" \ --env MSTY_TIMEOUT="Service timeout in seconds (default: 10)" \ --env MSTY_MLX_PORT="MLX service port (default: 11973)" \ --env MSTY_VIBE_PORT="Vibe CLI Proxy port (default: 8317)" \ --env ANTHROPIC_API_KEY="Bloom judge API key (required for Bloom phase)" \ --env MSTY_LLAMACPP_PORT="LLaMA.cpp service port (default: 11454)"
How to use
Msty Admin MCP is a Python-based server that coordinates Msty Studio Desktop tooling via the Model Context Protocol. It provides 36 tools spread across six phases, including foundational checks, configuration synchronization, multi-backend model orchestration, and Bloom-based behavioral evaluation for safety and reliability. The server can connect to multiple local backends (Ollama, MLX, LLaMA.cpp, and Vibe Proxy) through port-based service discovery, automatically wiring available tools to the MCP transport (stdio or HTTP). This makes it suitable for automating admin tasks, inspecting internal state, and conducting cross-backend prompts and evaluations with consistent interfaces. The included Bloom evaluation features allow you to test for behaviors like sycophancy, hallucination, and overconfidence, and then decide whether to hand off to Claude for specific task categories.
How to install
Prerequisites:
- Python 3.10+
- pip (comes with Python)
- Optional: git if you want to clone the repository
Install from PyPI (quick start):
pip install msty-admin-mcp
Run with default MCP transport (stdio):
msty-admin-mcp
Run with HTTP transport (streamable-http) support:
pip install msty-admin-mcp[http]
msty-admin-mcp --transport streamable-http
From source (clone-and-install in editable mode):
git clone https://github.com/M-Pineapple/msty-admin-mcp
cd msty-admin-mcp
pip install -e .
Configure environment (optional, shown below) and run as a Python module:
# Example: run via module (uses default stdio transport)
python -m msty_admin_mcp
Additional notes
Environment variables:
- MSTY_HOST, MSTY_AI_PORT, MSTY_MLX_PORT, MSTY_LLAMACPP_PORT, MSTY_VIBE_PORT control connection to local backends.
- MSTY_TIMEOUT sets the per-tool service timeout in seconds.
- ANTHROPIC_API_KEY is required for Bloom behavioral evaluation (Phase 6); ensure you have an active Anthropic Bloom key.
Common issues:
- Service discovery may fail if the local backends are not running or ports differ from defaults.
- Bloom features require a valid ANTHROPIC_API_KEY; without it, Bloom evaluation tools will not execute.
- If running HTTP transport, ensure you have the http extra installed and that port 8000 is not in use by another service.
Configuration tips:
- Keep MSTY_HOST reachable from the MCP server to allow instrumented health checks and data collection.
- The metrics SQLite database is stored by default at ~/.msty-admin/msty_admin_metrics.db and is auto-initialized on first use.
Related MCP Servers
claude-additional-models
Reduce Claude Desktop consumption by 10x - Integrate Google's Gemini or Z.ai's GLM-5 (744B params) with Claude via MCP for intelligent task delegation
omega-memory
Persistent memory for AI coding agents
mcp-prompt-optimizer
Advanced MCP server providing cutting-edge prompt optimization tools with research-backed strategies
grok-faf
First MCP server for Grok | FAST⚡️AF • URL-based AI context • Vercel-deployed
mcp-salesforce
🚀 Complete MCP (Model Context Protocol) server for Salesforce integration with Claude Desktop. Provides seamless OAuth authentication, universal CRUD operations on any Salesforce object.
temple-bridge
The Sovereign Stack: MCP server binding local AI capabilities with governance protocols. 100% local operation with memory, conscience, and recursive observation.