Gantry

Gantry

An MCP proxy and dashboard for Space Molt AI fleets that provides compound tools, guardrails, multi-agent coordination, and real-time monitoring.

Category
访问服务器

README

Gantry

Dashboard

Gantry is an MCP proxy and live dashboard for Space Molt AI fleets — handle guardrails, compound tools, multi-agent coordination, and real-time monitoring in one server.

What It Does

Space Molt is a text-based space MMO played entirely through MCP (Model Context Protocol) tools. Gantry sits between your AI agents and the game server, providing:

  • Compound toolsbatch_mine, travel_to, jump_route, multi_sell, scan_and_attack, and more. One tool call that handles a full multi-step sequence, tick waits, and error recovery.
  • PrayerLang — A tiny server-side DSL (spacemolt_pray) for bounded, deterministic scripts. Agents submit a program once and the proxy runs it to completion, short-circuiting on predicates. Inspired by prayer.rs.
  • Guardrails — Rate limiting, per-tool call limits, decontamination (strips hallucination keywords from agent output before it persists), forbidden word enforcement.
  • Multi-agent coordination — Fleet-wide sell deconfliction, fleet order injection into tool responses, agent signal routing.
  • Live dashboard — React/Next.js web UI with agent status cards, real-time tool call streams (SSE), galaxy map, analytics charts, and agent notes.
  • Pluggable auth — Local network bypass, Cloudflare Access JWT validation, or no auth for local-only use.

Quick Start

Prerequisites: Bun and a Space Molt account.

# 1. Install
git clone https://github.com/geleynse/gantry.git
cd gantry
bun install
cd server && bun install && cd ..

# 2. Set up your fleet directory
bun server/scripts/gantry-setup.ts ./my-fleet
# Edit my-fleet/gantry.json — configure agents
# Add credentials to my-fleet/fleet-credentials.json
# Copy and customize a prompt: cp examples/agent-template/system-prompt.md my-fleet/my-agent.txt

# 3. Start the server
FLEET_DIR=./my-fleet bun run server/dist/index.js

Open http://localhost:3100 in your browser to see the dashboard.

Configure Claude Code to use Gantry as its MCP server:

{
  "mcpServers": {
    "spacemolt": {
      "type": "http",
      "url": "http://localhost:3100/mcp/v2"
    }
  }
}

Then run an agent turn:

claude -p "You are my-agent, a Space Molt trader. Login and take your turn." \
  --mcp-config _data/mcp.json

See the server README for full configuration and deployment details.

Key Features

Compound Tools

Gantry exposes 9 compound tools that handle full multi-step game sequences:

Tool What it does
batch_mine Mine N times, wait for ticks, stop on cargo full
travel_to Undock, travel, dock in one call with POI name resolution
jump_route Multi-hop jump sequence with auto-refuel and arrival tick detection
multi_sell Sell multiple items sequentially, check demand first, deconflict with fleet
passenger_run Load waiting passengers by destination into free berths and plan the delivery route
scan_and_attack Full combat loop: scan, pick target, battle loop, auto-loot
loot_wrecks Scan for wrecks and salvage them
battle_readiness Check hull, fuel, ammo, and nearby threats before combat
flee Exit combat and travel to safety

Each tool accepts parameters via the MCP tools/call interface. See the server API docs for endpoint details.

Live Dashboard

Galaxy Map

The web dashboard provides:

  • Agent status cards with health scores, ship info, faction badges
  • Live tool call stream (SSE) with pending state tracking
  • Galaxy map (505 systems, faction colors, agent positions)
  • Analytics: cost, iterations, credits over time
  • Agent notes: diary, strategy, market intel

Multi-Agent Coordination

Fleet orders are injected into tool responses at zero tool-call cost. Sell deconfliction warns agents when another fleet member recently sold the same item at the same station. Signals and comms are routed through SQLite, not files.

Installation

Option A: Docker (recommended)

Requires Docker and Docker Compose. Bun is optional (only needed if you want to use the setup script).

git clone https://github.com/geleynse/gantry.git
cd gantry

# Create a fleet directory (use setup script if you have Bun, or copy from examples/)
mkdir -p _data
cp examples/gantry.json.example _data/gantry.json
# Edit _data/gantry.json with your agent config

# Build and run
docker compose up --build -d

Dashboard at http://localhost:3100. Override the port with GANTRY_PORT=3101 docker compose up -d.

Option B: From Source (Bun)

git clone https://github.com/geleynse/gantry.git
cd gantry
bun install
bun run build    # builds server + Next.js dashboard
bun run dev      # development mode with hot reload

Option C: Single Binary

For deployments without Bun or Docker — one ~200MB executable.

# Build (requires Bun locally)
cd gantry/server
bun install && bun run build:binary

# Scaffold fleet directory locally, then deploy
bun scripts/gantry-setup.ts ./my-fleet
# Edit my-fleet/gantry.json, add credentials and prompts

scp dist/gantry user@server:/opt/gantry/
scp -r my-fleet user@server:/opt/gantry/fleet

# Run on target (no Bun needed)
ssh user@server
FLEET_DIR=/opt/gantry/fleet GANTRY_SECRET="$(openssl rand -hex 32)" /opt/gantry/gantry

The binary is fully self-contained — static frontend assets are embedded at compile time.

Documentation

Architecture

Claude Code / Codex CLI
        │
        │  MCP (HTTP)
        ▼
Gantry Server :3100
  ├── /mcp/v2        MCP proxy (compound tools, guardrails, injections)
  ├── /api/*         REST API (agent status, comms, analytics, notes)
  └── /              Web dashboard (React + Next.js, SSE streams)
        │
        │  MCP (HTTP, optionally via SOCKS proxy)
        ▼
game.spacemolt.com/mcp

All agent data is stored in SQLite (fleet.db). The server is a single Express process running on Bun.

Tests

bun test          # run all tests

Contributing

See CONTRIBUTING.md.

License

MIT — see LICENSE.

推荐服务器

Baidu Map

Baidu Map

百度地图核心API现已全面兼容MCP协议,是国内首家兼容MCP协议的地图服务商。

官方
精选
JavaScript
Playwright MCP Server

Playwright MCP Server

一个模型上下文协议服务器,它使大型语言模型能够通过结构化的可访问性快照与网页进行交互,而无需视觉模型或屏幕截图。

官方
精选
TypeScript
Magic Component Platform (MCP)

Magic Component Platform (MCP)

一个由人工智能驱动的工具,可以从自然语言描述生成现代化的用户界面组件,并与流行的集成开发环境(IDE)集成,从而简化用户界面开发流程。

官方
精选
本地
TypeScript
Audiense Insights MCP Server

Audiense Insights MCP Server

通过模型上下文协议启用与 Audiense Insights 账户的交互,从而促进营销洞察和受众数据的提取和分析,包括人口统计信息、行为和影响者互动。

官方
精选
本地
TypeScript
VeyraX

VeyraX

一个单一的 MCP 工具,连接你所有喜爱的工具:Gmail、日历以及其他 40 多个工具。

官方
精选
本地
graphlit-mcp-server

graphlit-mcp-server

模型上下文协议 (MCP) 服务器实现了 MCP 客户端与 Graphlit 服务之间的集成。 除了网络爬取之外,还可以将任何内容(从 Slack 到 Gmail 再到播客订阅源)导入到 Graphlit 项目中,然后从 MCP 客户端检索相关内容。

官方
精选
TypeScript
Kagi MCP Server

Kagi MCP Server

一个 MCP 服务器,集成了 Kagi 搜索功能和 Claude AI,使 Claude 能够在回答需要最新信息的问题时执行实时网络搜索。

官方
精选
Python
e2b-mcp-server

e2b-mcp-server

使用 MCP 通过 e2b 运行代码。

官方
精选
Neon MCP Server

Neon MCP Server

用于与 Neon 管理 API 和数据库交互的 MCP 服务器

官方
精选
Exa MCP Server

Exa MCP Server

模型上下文协议(MCP)服务器允许像 Claude 这样的 AI 助手使用 Exa AI 搜索 API 进行网络搜索。这种设置允许 AI 模型以安全和受控的方式获取实时的网络信息。

官方
精选