Compare a SQL data layer with separate SaaS MCP tools
At a glance
A former PostHog AI builder tests Dinobase on business questions that join data from multiple sources. This workflow is listed under data-agents, text-to-sql, benchmarking, cross-source-analysis.
The builder synced source data to Parquet, exposed it through DuckDB and used a Claude agent to annotate schemas. He compared that SQL layer with separate per-source MCP access across 11 models.
How it went
He reports two to three times the answer accuracy and 16 to 22 times fewer tokens per correct answer in his experiment.
Worth knowing
The benchmark used 75 questions over synthetic data with an LLM judge. In replies, he also acknowledged stale columns after schema changes.
Save ideas in this browser. Upvotes count across the board.
An indie hacker packaged an agency's eight repeated workflows (briefs, client approval with revision counters, quotes with margin checks, content calendars, post-mortems) as free MIT Claude skills.
Anthropic’s Building Effective Agents appendix highlights two proven patterns—support agents that pull data and take actions like refunds, and coding agents that iterate against tests—with clear success metrics and human review.
Indie Hackers founder Sean Prentice used Claude Code to build an orchestrator + marketing director + specialist agents (content/ads/research) for his SaaS; outputs land in Slack and nothing posts or spends without his explicit say (~$10/mo self-use).