Skip to main content
Glama
ArtemKyslicyn

aichallenge-mcp-kit

by ArtemKyslicyn
OverviewSchemaSourceRelated ServersDiscussionsAdmin
Python
Local

aichallenge-mcp-kit

Hierarchical MCP hub for your machine. One Streamable HTTP /mcp + Bearer token. Attach child MCPs (filesystem, git, montage, …), run sandboxed Python, move files via /artifacts, then paste the tunnel URL into AIChallenge → Свой MCP.

AIChallenge never runs these processes. This kit never talks to the site database. The site only sees one URL + one token.

Architecture: chat → tunnel → hub → groups / artifacts

Why a separate public repo?

Site

Guest MCP client (SSRF-safe, chat tools)

This kit

Your PC: hierarchy, sandbox, big files

Stdio MCP servers cannot be reached from a cloud chat. The kit publishes them behind one HTTPS tunnel.

Related MCP server: ChatGPT Desktop Commander MCP

5-minute connect

Connect: clone → run → tunnel → Guest MCP

git clone https://github.com/ArtemKyslicyn/aichallenge-mcp-kit.git
cd aichallenge-mcp-kit
cp .env.example .env          # set KIT_SHARED_TOKEN (long random string)
cp kit.example.yaml kit.yaml  # enable the children you want
uv sync
uv run aichallenge-mcp-kit    # listens on 127.0.0.1:3100

In another terminal:

cloudflared tunnel --url http://127.0.0.1:3100

On the site (logged in): Свой MCP / Настройки → Подключения:

  • URL: https://<tunnel-host>/mcp

  • Token: same value as KIT_SHARED_TOKEN

Optional pack: packs/aichallenge-guest.example.json (replace the URL; never put the token in a shared pack file).

Full walkthrough: docs/connect-aichallenge.md.

What you get

  • Hierarchy: kit.yaml → groups → children. Tool names: {group}__{child}__{tool} (e.g. dev__python__exec).

  • Hub tools: hub_list_children, hub_workspace_list|put|get.

  • Python sandbox: dev__python__* — cwd under workspace/sandboxes/, network off by default.

  • Stdio / HTTP proxy: enabled children stay connected for the life of the process; tools are forwarded with prefixes.

  • Artifacts: large outputs via GET /artifacts/{id} with the same Bearer.

Python sandbox

Docs

Doc

Topic

docs/architecture.md

Hub, groups, naming

docs/connect-aichallenge.md

Tunnel + Guest MCP

docs/python-sandbox.md

Exec limits

docs/children.md

Adding stdio/HTTP children

docs/security.md

Tokens, sandbox, risks

Design origin (AIChallenge monorepo): docs/superpowers/specs/2026-09-28-mcp-kit-hub-design.md.

Requirements

  • Python 3.12+

  • uv recommended

  • cloudflared (or ngrok) for a public HTTPS URL

  • Optional: Node/npx for official filesystem MCP; uvx for git MCP

Security (read this)

Whoever has URL + Bearer can drive every enabled child and the Python sandbox on your machine (within workspace limits). Bind stays on loopback; only the tunnel is public. See docs/security.md.

License

MIT — see LICENSE.

This server cannot be deployed

One endpoint for every MCP toolRayrun runs every MCP server behind one URL and holds the credentials. Each call is decided per client, tool, and argument, then recorded.ray.runAd

Maintenance

ActivityMaintained
ResponsivenessNo issues

Resources

Related MCP Connectors

Related MCP Servers

  • Levitate

    iomz
    A
    license
    Not graded
    quality
    A
    maintenance
    Lifts local stdio MCP servers into remote Streamable HTTP endpoints for cloud-hosted AI clients, with bearer-token auth and tool policy filtering.
    9 npm
    MIT

  • ChatGPT Desktop Commander MCP

    kathirm1323-ai
    F
    license
    Not graded
    quality
    B
    maintenance
    A Python MCP server that allows ChatGPT to execute commands on your local PC via a secure Cloudflare tunnel.
    -

  • mcp-chat-bridge

    RabbitsXx
    A
    license
    Not graded
    quality
    C
    maintenance
    Exposes local stdio MCP servers to chat AIs through temporary public URLs, providing both a Streamable HTTP gateway and a REST+OpenAPI bridge for integration with ChatGPT and other clients.
    MIT