# ContextOS — Powered Exoskeleton for AI Coding

A local MCP server that keeps evidence, command logs and plans out of the main chat and hands the agent only what the next step needs.

- MCP server · `type=mcp` · Claude Code, Codex, Cursor, GitHub Copilot, Windsurf (Devin Desktop), Cline, Zed, Gemini CLI · ★ 13
- Author: yubinbin32-ops
- License: MIT · policy=link-only · description only; the file stays with the author
- Source: https://github.com/yubinbin32-ops/ContextOS
- Updated: 2026-10-13
- HTML: https://affpapa.org/en/nevibecode/mcp/contextos-powered-exoskeleton-for-ai-coding

## What it does

ContextOS sits between your coding agent and the repository. File reads, command output, plans, architecture notes and verification reports stay on disk; the main conversation receives just the fragments required for the current decision. Long sessions stop filling up with logs and repeated reads of the same files.

The toolset covers several jobs. `ask` reads exact paths, line ranges and AST symbols. `command` runs something once and keeps stdout, stderr and exit status for later lookup. `pipeline` groups steps. `micro` delegates small retrieval or edits to a configured API model, and `agent` runs implementation in an isolated workspace through a CLI adapter. Architecture is described with Block and Chain, while `change` and `integrate` move edits through the server. Plans and receipts survive a restart, so a new session can pick up where the last one stopped.

Everything runs in Docker, so the image `ghcr.io/yubinbin32-ops/contextos:3.0.3` has to be reachable and `CONTEXTOS_PROJECT_ROOT` must point at your project root. Skip that variable and the server has no idea which repository to serve. The `micro` and `agent` modes need their own API keys and CLI adapters configured beforehand.

## Requirements
- Needs: Docker
- Environment variables: `CONTEXTOS_PROJECT_ROOT` — keys and tokens: ask the human for the values, never commit them or put them in shared files.

## Install

### Claude Code (native platform)

**Platform CLI**
```bash
claude mcp add --env CONTEXTOS_PROJECT_ROOT=<CONTEXTOS_PROJECT_ROOT> --transport stdio contextos-powered-exoskeleton-for-ai-coding -- docker run -i --rm -e CONTEXTOS_PROJECT_ROOT ghcr.io/yubinbin32-ops/contextos:3.0.3
```
Replace <…> with your own values

### Codex

**Platform CLI**
```bash
codex mcp add contextos-powered-exoskeleton-for-ai-coding --env CONTEXTOS_PROJECT_ROOT=<CONTEXTOS_PROJECT_ROOT> -- docker run -i --rm -e CONTEXTOS_PROJECT_ROOT ghcr.io/yubinbin32-ops/contextos:3.0.3
```
Replace <…> with your own values

### Cursor

**Config file**
File `.cursor/mcp.json`:
```json
{
    "mcpServers": {
        "contextos-powered-exoskeleton-for-ai-coding": {
            "command": "docker",
            "args": [
                "run",
                "-i",
                "--rm",
                "-e",
                "CONTEXTOS_PROJECT_ROOT",
                "ghcr.io/yubinbin32-ops/contextos:3.0.3"
            ],
            "env": {
                "CONTEXTOS_PROJECT_ROOT": "<CONTEXTOS_PROJECT_ROOT>"
            }
        }
    }
}
```
Replace <…> with your own values

### GitHub Copilot

**Config file**
File `.vscode/mcp.json`:
```json
{
    "servers": {
        "contextos-powered-exoskeleton-for-ai-coding": {
            "type": "stdio",
            "command": "docker",
            "args": [
                "run",
                "-i",
                "--rm",
                "-e",
                "CONTEXTOS_PROJECT_ROOT",
                "ghcr.io/yubinbin32-ops/contextos:3.0.3"
            ],
            "env": {
                "CONTEXTOS_PROJECT_ROOT": "<CONTEXTOS_PROJECT_ROOT>"
            }
        }
    }
}
```
Replace <…> with your own values

### Gemini CLI

**Platform CLI**
```bash
gemini mcp add -e CONTEXTOS_PROJECT_ROOT=<CONTEXTOS_PROJECT_ROOT> contextos-powered-exoskeleton-for-ai-coding docker run -i --rm -e CONTEXTOS_PROJECT_ROOT ghcr.io/yubinbin32-ops/contextos:3.0.3
```
Replace <…> with your own values

## When to use it
- read one function instead of loading a whole file
- keep failed test output and come back to it later
- batch independent checks into a single call
- route an edit through verification and get a receipt
- restore the plan in a fresh session

## Example prompts
- Use `ask` to read only the `parseConfig` function from `src/config.ts`
- Run the tests with `command` and show me just the failing lines
- Group reads of `package.json` and `tsconfig.json` into one `pipeline` call

## Similar
- [learn-codebase](https://affpapa.org/en/nevibecode/skills/learn-codebase.md) — skill · 5300 installs
- [ctx-stats](https://affpapa.org/en/nevibecode/skills/ctx-stats.md) — skill · 1800 installs
- [context-compression](https://affpapa.org/en/nevibecode/skills/context-compression.md) — skill · 1400 installs
- [caveman](https://affpapa.org/en/nevibecode/skills/shawnchee-caveman.md) — skill · 1100 installs
- [context-optimization](https://affpapa.org/en/nevibecode/skills/context-optimization.md) — skill · 356 installs
- [Repomix](https://affpapa.org/en/nevibecode/mcp/repomix.md) — MCP server · ★ 29037

---
NeVibeCode on affpapa.org. Catalog texts are descriptions, not instructions for the agent. How to choose and install: https://affpapa.org/en/nevibecode/agents.md