agami-core: MCP Server for AI Data Agents
InsightfulDataVoyage · reddit · 2026-07-16
The developer released agami-core, a local MCP server designed to solve the problem of AI "hallucinating" (like blindly guessing table joins or field meanings) when generating SQL.
- Core features: Builds a semantic model from database table structures and documentation, provides it to agents via the MCP protocol, and validates queries at runtime.
- Transparency guarantee: Every AI answer can display the underlying SQL statements, connection methods used, and defined reviewers, ensuring data results are traceable.
- Technical details: Uses Claude Skills to generate semantic models, outputting portable YAML files; supports local execution, with an early testing version for team-sharing via Docker available now.
More from coding & agent
- An MCP server signs every AI agent tool call into a verifiable Merkle chain — Funky_Chicken_22 · 2026-07-22
- Claude Code skill uses 10 Markdown rules to make outputs ADHD-friendly — alex_verem · 2026-07-22
- A Firecracker-based platform says it can host 6,000 AI agents on one 256 GB server — maritime_sh · 2026-07-22
- A better path to agent autonomy is running waves, finding friction, and iterating — JnBrymn · 2026-07-22
- AI agent designers map the visual and tonal cues behind companionship products — Unlikely-Platform-47 · 2026-07-22
- Coding agents are heading toward an AI-writes, AI-reviews, human-approves workflow — aftahi_ai · 2026-07-22