Ask any AI about the Stellar ecosystem: Lumen Loop's MCP and skills are in beta
Lumen Loop's MCP server is in beta. Any AI assistant with MCP support, including ChatGPT, Claude, and Gemini, can query the Stellar ecosystem corpus for free: 756 projects, 912 SCF submissions, articles, talks, and events. A companion set of Claude skills automates common research workflows.

The Lumen Loop MCP server is in beta. The same indexed Stellar ecosystem that powers lumenloop.com is now a read-only endpoint at https://mcp.lumenloop.com that any AI assistant with MCP support, including ChatGPT, Claude, and Gemini, can connect to for free. A companion set of skills for Claude ships alongside it.
You could already point an AI at lumenloop.com and have it read a page. The MCP changes what that is worth. Instead of guessing URLs, a model searches the whole corpus by meaning, and it reaches structured data that crawling never surfaced well: the Stellar Community Fund archive, and every project's cross-referenced footprint.
What sits behind the endpoint
lumenloop.com folds the ecosystem into one cross-referenced corpus: 756 projects in the directory, 912 Stellar Community Fund submissions, and thousands of articles, talks, and events, each linked back to the project it concerns. The MCP serves that corpus as read-only query tools. Every result is the public, AI-summarized surface of the site. The tools never write, and they only return what is already published.
The questions it answers
Connected, a model can do the digging that used to mean a dozen open tabs and a lot of guessed URLs.
- Search by meaning. Semantic search runs across the indexed articles and audio/video, so "confidential payroll on Soroban" surfaces the right talk even when the words do not match.
- Check an SCF idea. Search the Community Fund archive by topic or by similarity to an existing submission, and see what has already been funded before you write the application.
- Vet one project. Ask for a single project's footprint and you get its description, category, tags, social handles, who has written about it, the talks that mention it, and its SCF history in one pass.
- Map a sector. "What RWA projects are live on Stellar, and who has covered them?" returns the directory matches plus the content around each name.
Connecting
The server speaks standard MCP, so any client with connector support works the same way: point it at https://mcp.lumenloop.com, sign in once, and the tools appear in the session. ChatGPT, Claude, and Gemini all support this today. Per-client setup, including the Claude Code one-liner, is at lumenloop.com/ai.
Skills turn tools into workflows
Raw tools are half of it. A connected model still has to know which tool to reach for, in what order, for a given job. For people working in Claude, lumenloop-skills are playbooks that encode those sequences. Other clients call the same tools directly; the skills are the Claude-side shortcut.
Seven ship in the first set. stellar-ecosystem-scout maps a sector or topic into a landscape of projects and the content around them. stellar-project-dossier builds a due-diligence profile of one project. scf-submission-radar checks an SCF idea against prior submissions before you apply. stellar-integration-finder shortlists the right wallet, oracle, anchor, RWA issuer, or DEX to integrate. stellar-ecosystem-digest produces a dated, cited digest of recent activity on a theme or named entity. stellar-builder-quickstart goes from an idea to the Stellar primitives that fit and routes to a build path. lumenloop-mcp-connect connects the server and points you at the right skill.
Install them as a Claude Code plugin:
/plugin marketplace add lumenloop/lumenloop-skills /plugin install lumenloop-skills@lumenloop
That enables the skills and wires the connector in one step. The source is at github.com/lumenloop/lumenloop-skills.
What to expect from a beta
The corpus has gaps. Some projects are missing, some content is filed under the wrong project, some summaries are thinner than the work deserves. The MCP reads all of that back faithfully, rough edges included. It is read-only and projected: a model gets the public summaries and metadata, never raw article bodies, transcripts, or private fields, and it can report what the ecosystem is doing but cannot change anything in it. It indexes the public ecosystem, so it is not a substitute for a project's own docs or stellar.org.
Start here
Point any MCP client at https://mcp.lumenloop.com, or install the Claude skills and let them wire it for you. The walkthrough, the tool list, and the skill catalog are at lumenloop.com/ai.
Sources
- lumenloop.com/ai
- mcp.lumenloop.com
- github.com/lumenloop/lumenloop-skills