Real Threads is designed to be understood by machines as well as people. Three complementary layers make the site consumable by LLMs, AI agents, and developer tools — each serving a different audience and use case.
Together they mean that whether someone is chatting with the on-site AI Guide, connecting an external tool via MCP, or simply fetching /llms.txt, they get structured, up-to-date context about every project on the site.
Primers & the AI Guide
Every project on Real Threads has a primer — a structured markdown document that gives an AI agent the background it needs to discuss that project knowledgeably. Primers follow a consistent template (Origins, Structure, Key Concepts, Related Projects, Agent Guidance) and are classified into three tiers based on depth:
Full (~50–100 lines) for complex interactive projects like Energy and Vector Atlas. Standard (~25–50 lines) for projects with moderate conceptual vocabulary. Lightweight (~10–25 lines) for focused tools and documentation like this one.
The AI Guide loads primers on demand through its tool set. When a visitor asks about a project, the guide calls getPrimer to fetch the relevant primer and uses it to ground its response in accurate, curated context — no hallucination, no stale training data.
MCP Server
For third-party tools and external agents, Real Threads exposes a Model Context Protocol (MCP) server. MCP is an open standard that lets AI applications discover and consume structured context from any compatible service.
The server provides four tools for querying projects, retrieving primers, searching the content catalog, and fetching individual artifacts — plus two resources for site-level context. Any MCP-compatible client (VS Code, Cursor, Claude Desktop, custom agents) can connect and start exploring the site programmatically.
Full endpoint details, tool reference, and client configuration examples are on the MCP Server documentation page.
Agent-Facing Routes
Two route conventions make the site directly browsable by agents that don't use MCP:
Primer routes at /projects/[slug]/primer serve a rendered primer page for every registered project. These are standard HTML pages that any web-browsing agent can fetch and parse.
/llms.txt is a machine-readable index inspired by llmstxt.org. It lists every available primer grouped by tier, as markdown links. When a new primer is registered, it appears in /llms.txt automatically — no manual route changes needed. One honest caveat: it is served as text/plain, so its links are plain text. Some fetching agents build their followable-URL list from extracted HTML <a href> anchors and never treat those markdown links as reachable.
/primers closes that gap: it is an HTML index of the same primer set — built from the same registry, so the two never diverge — with a real anchor to every primer. One fetched URL makes the whole set discoverable to any agent that follows HTML links.