All projects

AI Engineering / Developer Tooling

Experimental

AI Developer Tooling & MCP Lab

A controlled laboratory for connecting coding agents and local models to developer tools and data services, with particular attention to MCP initialization, permissions, and tool calling.

MCPPostgreSQLSupabaseOpenCodeVS CodeLocal LLM
Abstract AI agent and MCP gateway connected to PostgreSQL and Supabase data services

Context

The Problem

Coding agents needed a controlled way to reach local models and database-backed developer tools without defaulting to unrestricted administrative database access.

System flow

Architecture

VS Code / OpenCode / Coding Agents
OpenAI-Compatible Local LLM
MCP Integrations
PostgreSQL
Supabase

Implementation

Engineering

  • Connected AI coding agents to local model endpoints and experimented with tool calling.
  • Integrated PostgreSQL and Supabase through MCP-based tooling.
  • Created dedicated PostgreSQL roles for AI tooling with controlled schema permissions.
  • Managed environment configuration for local models, developer tools, and MCP services.
  • Debugged MCP initialization and local-network connectivity.

What required care

Challenges

  • Granting useful database capabilities while avoiding unrestricted administrative access.
  • Diagnosing MCP initialization across multiple tools and local services.
  • Working through tool-calling limitations in local models.

Technology

Stack

Status: Experimental

MCPPostgreSQLSupabaseOpenCodeVS CodeLocal LLM