pente-mcp
Enables AI to play Pente live against a human via MCP, using ChanBlake's browser game as a shared board. Provides create, join, move, and view tools for real-time multiplayer matches.
README
pente-mcp — an AI/MCP add-on for playing Pente
An MCP server that lets any MCP-capable AI play Pente. It supports two transports with the same four tools:
- Local (default) — two agents on the same machine play each other with a self-contained game engine and a local TCP socket. No website, no PeerJS, no WebRTC, no public cloud, no internet.
- Web — play against the ChanBlake browser game / a human over the public PeerJS cloud (a human opens an invite URL in their browser).
This is an add-on / integration layer, not a fork of the game. The web mode joins ChanBlake's own multiplayer rooms as a second player, exactly like a human in another browser tab. None of ChanBlake's game code is included here — it stays in his repository. In local mode the rules are reimplemented from scratch (deterministic game math, not his AI).
| Tool | Role | What it does |
|---|---|---|
create(transport?) |
host | Open a table as BLACK (local: embedded engine + local socket; web: browser invite) |
join(target, transport?) |
guest | Join a table as WHITE (local: target = local port; web: target = room code) |
move(row, col) |
either | Play one stone, wait for the opponent reply |
view |
either | Dump the synchronized board + game state |
Two transports
Local (agent-vs-agent, same machine) — transport="local"
Two agents (e.g. Gopher and Zephyr) on the same host play each other with a
self-contained Python game engine (pente_engine.py) and a plain JSON-line TCP
socket (pente_local.py). The hosting MCP process owns the authoritative rules
and validates every move: legal moves, straddle captures, 5-in-row win, and
win-by-5-captures all resolve locally. No internet required.
# Agent A hosts (BLACK)
create(transport="local")
# -> { "transport": "local", "port": 9333, "i_am": "BLACK", ... }
# Agent B joins (WHITE) — same machine
join(target=9333, transport="local")
# -> { "transport": "local", "port": 9333, "i_am": "WHITE", "connected": true }
Web (browser / remote) — transport="web"
Connect to ChanBlake's browser game over the public PeerJS cloud, so a human (or another agent) playing in the browser is the opponent.
create(transport="web") # -> { "join_url": "https://chanblake.github.io/pente/?room=ABC234", ... }
join(target="ABC234", transport="web")
What's in here
pente_mcp.py— the stdio MCP server (create / join / move / view, both transports).pente_engine.py— pure-Python Pente rules (legal moves, captures, wins, pro-opening). Zero dependencies.pente_local.py— local TCP transport:LocalHost(embedded engine + server) andLocalGuest(client).pente_local_host_cli.py— run the local host engine as a JSON-lines subprocess.pente_scan.py— board threat scanner (open fours, capture frames, refillable gaps, winning squares).pente-bot.js/pente-host.js/webrtc-polyfill.js— Node adapters for the web transport (PeerJS/WebRTC).test/— headless tests (fake host + the local transport tests).docs/— the original design spec.
Requirements
- Local mode: Python ≥ 3.11 with
mcp(pip install mcp). No Node, no internet. - Web mode: Node.js +
@roamhq/wrtcshim + internet to the public PeerJS cloud (0.peerjs.com:443).
Quickstart
pip install mcp
python pente_mcp.py # stdio MCP server (local transport is the default)
Register the server in your MCP client, then:
create(transport="local") # host a local game, BLACK
view() # full 19x19 board, whose_turn, captures, game_over
move(row=9, col=9) # your_move (+ their move if the opponent replied)
Tests
python pente_engine.py # rules self-test
python test_local_rules.py # win + capture via the local host engine
python test_local_two_process.py # cross-process host + guest over TCP
python test_mcp_agent_vs_agent.py # two full MCP processes playing each other
All four must pass. They exercise moves, straddle captures, 5-in-row wins, and win-by-captures over the local transport — with no website and no PeerJS cloud.
Credit
The web mode connects to ChanBlake's Pente: https://github.com/ChanBlake/pente — an unofficial, non-affiliated integration layer. Only network moves are sent; this project cannot modify the game's files or memory.
Known limits
- Local mode supports one active table per MCP server process (single
_activehandle). - Web room codes: the app strips
I,Oand0.createonly emits codes from its alphabet. - Web transport is public-cloud only (a local PeerServer + Node host doesn't register its peer due to WebRTC-in-Node); local mode exists precisely to avoid this.
License
MIT — this integration layer. ChanBlake's game has its own license (see his repo).
Recommended Servers
playwright-mcp
A Model Context Protocol server that enables LLMs to interact with web pages through structured accessibility snapshots without requiring vision models or screenshots.
Magic Component Platform (MCP)
An AI-powered tool that generates modern UI components from natural language descriptions, integrating with popular IDEs to streamline UI development workflow.
Audiense Insights MCP Server
Enables interaction with Audiense Insights accounts via the Model Context Protocol, facilitating the extraction and analysis of marketing insights and audience data including demographics, behavior, and influencer engagement.
VeyraX MCP
Single MCP tool to connect all your favorite tools: Gmail, Calendar and 40 more.
graphlit-mcp-server
The Model Context Protocol (MCP) Server enables integration between MCP clients and the Graphlit service. Ingest anything from Slack to Gmail to podcast feeds, in addition to web crawling, into a Graphlit project - and then retrieve relevant contents from the MCP client.
Kagi MCP Server
An MCP server that integrates Kagi search capabilities with Claude AI, enabling Claude to perform real-time web searches when answering questions that require up-to-date information.
E2B
Using MCP to run code via e2b.
Neon Database
MCP server for interacting with Neon Management API and databases
Exa Search
A Model Context Protocol (MCP) server lets AI assistants like Claude use the Exa AI Search API for web searches. This setup allows AI models to get real-time web information in a safe and controlled way.
Qdrant Server
This repository is an example of how to create a MCP server for Qdrant, a vector search engine.