Skip to content

opensite-ai/opensite-skills / linux-ai-dev-stack

linux-ai-dev-stack

Flagged

Install and wire the AI toolchain on a Linux dev machine: local LLM inference (Ollama, llama.cpp, vLLM, SGLang, TabbyAPI) with VRAM-based model selection, AI coding agents and CLIs (Claude Code, Codex, aider, OpenCode) with MCP servers, ML frameworks (PyTorch/JAX, Jupyter, HuggingFace) and GPU containers, and cloud/remote GPU workflows (RunPod, Vast.ai, Lambda, remote dev over SSH and Tailscale). Use when someone wants to run local models, set up or connect AI coding agents, configure MCP servers, route agents to local or cloud models, install an ML/GPU Python stack, or offload work to a rented GPU. It runs on the machine and assumes working GPU drivers.

references/mcp-servers.md

View on GitHub
# MCP Servers

Model Context Protocol (MCP) connects AI apps to external systems. This file covers
what MCP is, the essential servers for a dev box, where config lives per client, and
— most importantly — the security model. Treat **every server as untrusted code and
a prompt-injection surface.** Server names/endpoints are **as of 2026-07 — verify.**

## 1. What MCP is

An open standard (open-sourced by Anthropic, now broad cross-vendor adoption) —
"USB-C for AI." A **client** (Claude Code, ChatGPT, VS Code, Cursor, Zed, Gemini
CLI, Goose…) connects to **servers** that expose **tools**, **resources**, and
**prompts**. There is a public **MCP Registry** (`registry.modelcontextprotocol.io`)
and per-vendor directories (e.g. `claude.ai/directory`).

**Transports:**
- **stdio** — local subprocess (most dev servers; direct system access). Default when
  a JSON entry has a `command`.
- **HTTP (`streamable-http`)** — remote/hosted servers. A JSON `url` entry **must**
  set `"type": "http"` (or `sse`/`ws`) or the client treats it as stdio and fails.
- **SSE** — deprecated; prefer HTTP.

## 2. Essential servers for a dev box

Reference servers (maintained by the MCP steering group) + widely used community/vendor:

- **filesystem** — scoped file operations. Scope it to specific dirs.
- **git** — read/search/manipulate local repos.
- **github** — issues/PRs/code search/CI (official GitHub MCP server).
- **context7** (`@upstash/context7-mcp`) — up-to-date, version-correct library docs
  injected on demand; remote `https://mcp.context7.com/mcp` (API key for higher
  limits) or local via npx.
- **playwright** — drive a real browser (nav/click/snapshot/console/network) for E2E.
- **memory** — knowledge-graph persistence across sessions.
- **fetch** — web page → markdown.
- **sequential-thinking** — structured multi-step reasoning helper.
- Also common: postgres/database, sentry, cloudflare, vercel, serena (semantic code),
  and hundreds more via the registry.

## 3. Configuring servers (per client)

**Claude Code** with scopes — **Local** (`~/.claude.json`, current project only,
private), **Project** (`.mcp.json`, committed, team-shared), **User** (all projects):
```bash
claude mcp add --scope user context7 -- npx -y @upstash/context7-mcp --api-key ${CONTEXT7_API_KEY}
claude mcp add --scope project filesystem -- npx -y @modelcontextprotocol/server-filesystem ./
claude mcp add --scope project git -- npx -y mcp-server-git
claude mcp add --transport http github https://api.githubcopilot.com/mcp/
claude mcp add --scope project playwright -- npx -y @playwright/mcp
claude mcp add-json <name> '{"type":"http","url":"...","headers":{...}}'
```
Tools are referenced as `mcp__<server>__<tool>` in permission rules / `allowed-tools`
/ hooks. Other clients each have their own location (`~/.gemini/settings.json`,
per-IDE settings, `opencode.json`, `.cursor/mcp.json`, VS Code `mcp.json`) but the
server-definition shape (command/args/env, or url/type/headers) is portable.

## 4. Security (the important part)

- **Every server is untrusted code + a prompt-injection surface.** Servers that
  fetch external content (web pages, issues, emails) can smuggle instructions into
  your agent. Only connect servers you trust; read the source of stdio servers you
  run locally.
- **Least privilege.** Scope filesystem servers to specific dirs; give tokens the
  minimum scopes; avoid mixing high-privilege write tools with content-fetching
  servers in the same session (a classic injection escalation path).
- **Deny/allow rules.** In Claude Code, `permissions.deny` can block `mcp__*` (all
  MCP) or specific tools; servers can mark tools `requiresUserInteraction` to force a
  prompt even in auto modes.
- **Secrets.** MCP env/headers often carry API keys — inject via a secrets manager,
  never commit `.mcp.json` with plaintext tokens (use env-var interpolation like
  `${GITHUB_TOKEN}`). See secrets guidance in [ai-coding-agents.md](ai-coding-agents.md).
- **Context/token bloat.** Many servers add lots of tools; use on-demand tool loading
  (e.g. `ENABLE_TOOL_SEARCH` in Claude Code) and prune servers you do not use.

## 5. Checklist before adding a server

1. Do I trust the publisher / have I read the source?
2. Is it scoped to the minimum filesystem paths and token permissions?
3. Am I keeping content-fetching servers isolated from high-privilege write tools?
4. Are all secrets injected from a manager (no plaintext in committed config)?
5. Is the correct scope chosen (Local for private, Project for team-shared)?
6. For HTTP servers, did I set `"type": "http"` explicitly?
reverse-engineer-anythingmorluto/reaReverse engineer native, managed, Electron/JavaScript, packaged, firmware, and browser targets with REA, and build agent environments (AgentEnv bundles) of web applications from a live instance or a capture, and of Linux desktop apps with a task from a request. Use shipped-artifact or requested runtime evidence to explain features, compare versions, decompile code, or guide a reconstruction. Skip REA for ordinary source-repository architecture analysis.Stars: 470kUpdated Claude CodeCodexVerifiedInstall
diagnosing-superpowersobra/superpowersUse when a superpowers session went wrong and your human partner wants to know why — repeated work, ignored plans, stumbles, poor results, a skill that didn't fire, "it took too long", "why is it so expensive", "what is it doing" — or wants to build a bug report for the superpowers maintainers, for the current session or a past one identified by id or path, on any harness.Stars: 322kUpdated Claude CodeCodexVerifiedInstall
dispatching-parallel-agentsobra/superpowersUse when facing 2+ independent tasks that can be worked on without shared state or sequential dependenciesStars: 322kUpdated Claude CodeCodexVerifiedInstall
executing-plansobra/superpowersUse when executing an implementation plan in the current session as the implementer yourself — your human partner chose inline execution, or no subagent tool is availableStars: 322kUpdated Claude CodeCodexVerifiedInstall