Hacker News: slava_

Show HN: DataFrey – MCP server for Snowflake with text-to-SQL agent

slava_ — Tue, 21 Apr 2026 19:01:49 +0000

I’m a data scientist and I find it hard to use Claude Code for SQL - it doesn’t have DB context. so I made yet another database MCP server! only Snowflake support for now.

I had to reconnect with nature after reading native Snowflake MCP setup docs so for my server I’ve made a nice CLI wizard to set up DB connection and install the Claude Code plugin: MCP + skill - you can ask it like `/db write dbt model to rank leads`.

It also has a `plan` tool for complex questions. when you ask a blurry question, it triggers a separate text-to-SQL agent that uses 1. (kinda) RAG for your schema (along with some values) that builds during DB connection (if you agree) 2. subagents to explore your data. 3. planning. This is what Snowflake Cortex is supposed to do, but when I try it, it never finds the right tables.

Database-as-MCP sounds like a security nightmare, but I put a lot of effort into making it safer. I’d appreciate any thoughts on the secure design. by default, CLI asks for select permissions on all schemas, not just information_schema. I’m convinced that it’s impossible to write good SQL without peeking into the data. maybe it's a hot take - share your thoughts!

Everything is free and hosted by me, but rate-limited. In the future, I want to charge for planning calls above the limit. I have a bunch of ideas on how to make a smarter text-to-SQL, so I want to keep this part closed-source. I’ll open-source more though - it’s just deployed as a monolith now.

Comments URL: https://news.ycombinator.com/item?id=47853007

Points: 1

# Comments: 0

New comment by slava_ in "Show HN: Marky – A lightweight Markdown viewer for agentic coding"

slava_ — Mon, 20 Apr 2026 16:32:22 +0000

that's really cool! I don't think having this in a separate window adds value. I feel like it's pretty conventional to open something for an agent in the browser. even for the agent itself. for example, ultraplan.

New comment by slava_ in "Yann LeCun says Dario Amodei "knows nothing about AI effects on jobs""

slava_ — Mon, 20 Apr 2026 16:18:51 +0000

I don't understand why debates on "AI effect on economy" are switching to "how good AI will be". It's not about "is technology capable" but about market response. and it's so random and unpredictable, so heavily depends on the sentiment, economic conditions, regulations.

New comment by slava_ in "Stop trying to engineer your way out of listening to people"

slava_ — Mon, 20 Apr 2026 16:04:04 +0000

I think it says a lot about micromanaging practice broadly. the person assigned for a task should be fully trusted and accountable. the manager could've criticize writing, recommend using LLM, but not interfere. what they've done shows lack of trust and responsibility first.

New comment by slava_ in "Show HN: Stage – Putting humans back in control of code review"

slava_ — Mon, 20 Apr 2026 15:35:05 +0000

did you think about this tool not from "review someone's PR" but from "review Claude's output" perspective? let's say, I work locally, /effort max and I get a huuuge diff. and then I run a skill "/turn-summary" and it walks me through the changes locally or in some nice UI opened in the browser and helps me to focus my attention in groups. or it would be nice native feature in tools like Conductor.

New comment by slava_ in "Show HN: Made a highly organised email client with prompt-free AI within"

slava_ — Mon, 20 Apr 2026 15:20:04 +0000

do I understand it correctly: you (1) have 30 pre-defined classes of emails, (2) you predict a class, (3) put email in a specific UI? I think dynamic clusters would be more helpful! every inbox is different. what if you would cluster my inbox and predict a cluster for a new email? you can do that based on the content and interactions.