mcporbital-space-research — Claude Skill
Drop one folder into Claude Desktop, Cursor, or claude.ai with Skills. Pair it with the OSS MCP server and Claude gets 5 grounded space-research tools with provenance discipline baked in — every claim cited, every numeric value hyperlinked to a primary source.
What ships
mcporbital-space-research/
├── SKILL.md
├── README.md
├── references/
│ ├── tool_catalog.md
│ └── safety_policy.md
├── scripts/
│ ├── verify_mcp.sh
│ ├── refusal_message.md
│ └── mission_brief.md.tpl
└── examples/
└── 01_apophis_brief.mdMostly markdown + a single bash health-check. The skill is Apache-2.0 and reviewable line-by-line — no opaque binary blobs.
Two commands. Two minutes.
# 1. Drop the skill folder where Claude looks mkdir -p ~/.config/claude/skills git clone https://github.com/lysechko1/mcporbital.com-claude-skill \ ~/.config/claude/skills/mcporbital-space-research # 2. Confirm the MCP server is reachable chmod +x ~/.config/claude/skills/mcporbital-space-research/scripts/verify_mcp.sh ~/.config/claude/skills/mcporbital-space-research/scripts/verify_mcp.sh # expected: "✓ @mcporbital/space-intelligence-mcp reachable. 5 tools advertised."
The verify script exercises the stdio handshake and counts advertised tools — gives a one-line green / red answer.
claude_desktop_config.json
macOS: ~/Library/Application Support/Claude/ · Windows: %APPDATA%\\Claude\\
{
"mcpServers": {
"space-intelligence": {
"command": "npx",
"args": ["-y", "@mcporbital/space-intelligence-mcp@latest"]
}
}
}Restart Claude Desktop after editing. Ask "give me a research brief on 99942 Apophis"— Claude chains search_space_object → get_object_profile → get_asteroid_close_approaches automatically.
When the skill activates
- research <object>
- brief on <object>
- what's near earth
- next pass of <satellite>
- space weather
- kp index
- TLE of
- ephemeris of
- papers about <topic>
- is <asteroid> dangerous
- mission concept
- close approach
- was there a storm on <date>
- + Russian variants — космический объект · астероид · бриф · и т.д.
Same policy as the MCP server
Research / simulation / decision-support / human-in-the-loop only. The skill refuses live commanding, deception, dual-use targeting; the MCP server enforces the same regex patterns server-side so even a miscoded handler can't bypass them.