mdflow — Context OS for AI Coding Agents MCP server

ContextOS — AI Coding Context OS Auto-governing context via MCP · Reducing context window waste by 90%+ in practice Stopping context explosion and hallucination in large codebases: surgical AST read/write instead of dumping whole files, out-of-context command receipts, and an intuitive Metro Map architecture graph.

⬇️ Download macOS App · ⚡ Start in 30 seconds · 中文文档

What problem does ContextOS solve?

ContextOS acts as a high-precision powered exoskeleton for AI coding assistants.

In traditional workflows with large repositories, an AI agent is forced to haul massive raw files, noisy build dumps, and repetitive documentation in its prompt. It quickly runs out of breath — the context window explodes, hallucinations multiply, and previous design decisions are forgotten.

With ContextOS, the AI sheds the deadweight and lets the MCP protocol auto-govern context:

In real-world multi-step tasks, ContextOS reduces redundant context consumption by over 90%, eliminating context explosion and memory loss across long-running sessions.

What changes in daily development?

1. Architecture becomes a Metro Map

Blocks bind to real files, AST symbols, or directory trees (zero ghost blocks allowed). Dependency and resource directories use one bounded tree anchor instead of per-file bookkeeping; Chains represent horizontal subway rails, and typed Links connect transfer stations orthogonally.

2. The agent states intent, the OS runs the loop

The agent only calls explore → change → verify → ship. The OS derives the working set from git, files code under AST-derived modules, attaches verification receipts automatically and keeps every response inside a context budget. Coverage and evidence gates are advisory by default; strict: true in .contextos/profile.json restores hard enforcement.

3. Commands run out-of-context

The command gateway behind verify and ops strips ANSI noise, redacts secrets, saves the full sanitized log into .contextos/logs/, and returns a compact receipt with critical error diagnostics, reducing terminal noise by over 98%.

4. Code tools operate surgically

Shortened. The full README is on GitHub.