routermcp
← Back to Blog

ML/AI Engineer MCP Pack (2025) — Hugging Face, LlamaIndex, Snowflake

• By RouterMCP Team

Experiment, evaluate, and deploy with MCP servers for model catalogs, retrieval, and warehouse AI. Includes quickstarts and evaluation tips.

ML workspace connecting HF models, retrieval utilities, and warehouse AI.

ML/AI Engineer MCP Pack (2025) — Hugging Face, LlamaIndex, Snowflake

TL;DR: Browse models/datasets (Hugging Face), wire retrieval with LlamaIndex MCP tools, and run SQL/Cortex pipelines in Snowflake — all via one endpoint.

Servers

  • Hugging Face MCP (official). https://github.com/huggingface/hf-mcp
  • LlamaIndex MCP Toolbox (community). https://github.com/stevengogogo/llamaindex_mcp
  • Snowflake MCP (Labs). https://github.com/Snowflake-Labs/sfguide-building-snowflake-mcp-server

Flow

  1. Pull candidate models and licenses from HF; record decisions.
  2. Create a retrieval graph with LlamaIndex MCP tools.
  3. Run batch evaluation or a Cortex function in Snowflake.

Links

  • Pack docs: /packs/ml-engineer
  • Related posts: Benchmarking (07), Observability (10)

FAQ Q: Can I run private eval datasets safely?
A: Keep data in‑region and restrict tool egress; see Compliance (12).

Schema

Checklist (fast)

  1. Intent. 2) Title/meta. 3) Slug. 4) TL;DR. 5) Flow. 6) FAQ. 7) Links. 8) Images/alt. 9) Edit. 10) CTA.

CTA

  • Use the template: examples/packs/ml-engineer.mcp.json and the eval harness + retrieval graph starter.