Show and tell: DeCodifier for deterministic, low-token Codex/MCP repo workflows

[ftrou]

I wanted to share a project I've been building: DeCodifier, a local-first coding engine and MCP-friendly tool layer focused on deterministic repo retrieval for AI coding agents.

The problem I'm trying to solve is predictable context selection when agents operate on larger codebases. Instead of jumping straight to full-file reads, the workflow is metadata-first:

• repo brief / stack + entrypoints
• task-aware read plan
• materialized context bundle under a token budget
• narrow line-range reads only when needed

Current capabilities include structural symbol extraction, lexical fallback search, file operations, starter packs, and a context sidecar for cached repo intelligence.

A few questions for people using Codex + MCP in practice:

• Which repo-context patterns have worked best for you?
• Would you rather have thinner planning tools or richer server-side context bundles?
• What MCP affordances are still missing for real coding workflows?

If useful, I can also share a minimal Codex/MCP setup example around this.