LumenLoop Stellar Skills

Preconfigured playbooks that make any MCP-capable AI assistant useful for researching and building on the Stellar ecosystem — wired to LumenLoop's free, read-only ecosystem MCP at https://mcp.lumenloop.com.

LumenLoop indexes the public Stellar ecosystem — projects, Stellar Community Fund (SCF) submissions, and editorial / news / audio-video content — and exposes it as 18 read-only query tools over the open Model Context Protocol. Any MCP client can connect (Claude, ChatGPT, Gemini, Cursor, …). This repo adds the playbooks for using those tools well, plus an .mcp.json that wires the connector.

The skills are packaged as Claude Code Agent Skills for one-command install, but the connector and the recipes inside each skill are model-agnostic — see Using these with other LLMs.

Read-only and public by design. These skills only ever query LumenLoop's public ecosystem data — they never write anything.

See VISION.md for the design and authoring standard.

/plugin marketplace add lumenloop/lumenloop-skills
/plugin install lumenloop-skills@lumenloop

Installing the plugin enables all eight skills and wires the lumenloop MCP connector from the bundled .mcp.json. Then connect the server (below).

git clone https://github.com/lumenloop/lumenloop-skills.git

Run Claude Code from the clone (or copy .mcp.json + skills/ into your project's .claude/). Project skills and the MCP server are auto-discovered.

git clone https://github.com/lumenloop/lumenloop-skills.git
cp -r lumenloop-skills/skills/* ~/.claude/skills/        # or symlink individual folders

Then add the server to your MCP config once:

claude mcp add --transport http lumenloop https://mcp.lumenloop.com

The skills query the lumenloop MCP server. After install, connect it in your client:

• Claude Code / desktop: the lumenloop server appears from .mcp.json (or the claude mcp add above). Run /mcp, pick lumenloop, and complete the one-time sign-in.
• Claude.ai, ChatGPT, Gemini & other MCP clients: add a custom / remote MCP connector pointing at https://mcp.lumenloop.com (transport: streamable HTTP) and sign in once.

Once connected, ask your assistant something like "map the RWA projects on Stellar" or "vet project <slug> for me" and the matching skill (or playbook) takes over. In Claude Code you can also invoke one directly, e.g. /stellar-ecosystem-scout real-world assets.

The /plugin install is Claude Code-specific, but nothing else here is:

1. Connect the MCP at https://mcp.lumenloop.com from any MCP-capable client (Claude, ChatGPT, Gemini, Cursor, or your own agent). The 18 tools work the same everywhere.
2. Use the skill files as playbooks. Each skills/<name>/SKILL.md is plain markdown describing exactly which tools to call, in what order, with which parameters. Paste one in as a system prompt / instructions, or hand it to your agent as a reference — the recipes are model-agnostic MCP tool-call sequences.

node scripts/validate_skills.mjs          # validate all skills + manifests
node scripts/validate_skills.mjs skills/stellar-ecosystem-scout   # one skill
claude plugin validate .                  # Anthropic manifest validation

The validator enforces the authoring contract in VISION.md, including the rule that skills may reference only the 18 public read tools. New skills are added one validated folder at a time.

MIT — see LICENSE. Stellar ecosystem data is served by LumenLoop and remains subject to its source terms.