How AI reads GitHub

github.com Jun 22, 2026 6 min read Not Ready

Short

18/ 100

AEO Level 1Not Ready

The short answer

GitHub earned an AEO score of 18 out of 100, placing it at level 1. Despite being the world's largest code-hosting platform with a robust REST and GraphQL API, the site provides almost no structured signals for AI agents to discover, authenticate against, or interface with its capabilities.

What AI sees

An AI agent landing on GitHub today encounters a JavaScript-heavy interface built for human developers, not machine readers.

When an AI crawler hits GitHub's homepage, it receives a dense HTML document populated with repository cards, trending projects, and marketing copy — but none of it is wrapped in machine-readable structure that guides an agent toward the platform's actual value. There are no Link response headers pointing to API endpoints, no MCP Server Card advertising GitHub's considerable automation surface, and no Content-Signal directives signaling how the content may be used. Structured data coverage lands at a modest 40/100, but identity and agent-interface layers score zero. The platform millions of developers rely on for automation is ironically invisible to the very AI agents that could orchestrate it.

GitHub homepage as an AI agent sees it — content-rich for humans, sparse on machine-readable agent signals

GitHub robots.txt: no AI crawler rules, no Content-Signal directives, no Sitemap reference

Where it loses points

Agent interfaces and identity/auth are the deepest gaps, both scoring 0 out of 100.

How to fix it

Three targeted server-side additions would dramatically lift GitHub's AI readiness without touching a line of application code.

Publish an API Catalog

Goal

Expose GitHub's REST and GraphQL APIs through a machine-discoverable catalog at /.well-known/api-catalog following RFC 9727.

Issue

The scan found no /.well-known/api-catalog endpoint, leaving AI agents with no standardized way to locate or enumerate GitHub's API surface.

Fix

Serve /.well-known/api-catalog as application/linkset+json with linkset entries for the REST API, GraphQL endpoint, and webhook documentation. Any RFC 9727-aware agent can then instantly enumerate every automation surface GitHub offers without reading a single doc page.

Add an MCP Server Card

Goal

Publish a Model Context Protocol Server Card so AI agents can discover GitHub's tooling capabilities and transport endpoint.

Issue

No /.well-known/mcp/server-card.json exists, which directly drives the agent_interfaces score of 0.

Fix

Create /.well-known/mcp/server-card.json declaring serverInfo, the MCP transport endpoint, and capabilities such as repository search and pull-request management. This single file turns GitHub into a first-class tool for any MCP-compatible AI agent without API changes.

Declare AI Crawler Rules in robots.txt

Goal

Add explicit per-user-agent directives for major AI crawlers so they know exactly which paths are available for indexing and citation.

Issue

GitHub's robots.txt contains no rules for GPTBot, ClaudeBot, PerplexityBot, or similar crawlers, leaving agent_discovery stranded at 25/100.

Fix

Add individual User-agent blocks for each major AI crawler with Allow directives covering public repository content and README files. Pair each block with a Content-Signal directive declaring ai-train and ai-input preferences so AI systems act on explicit permission rather than guessing at intent.

Common questions

Why does GitHub score so low on AEO despite having a well-documented API?

GitHub's API documentation is excellent for human developers who already know where to look, but AEO measures machine-discoverable signals — RFC 9727 API catalogs, MCP Server Cards, and robots.txt AI directives. None of those exist today, so AI agents cannot find or use GitHub's capabilities through standard discovery protocols even though the underlying API is genuinely powerful.

Would adding AEO signals affect GitHub's existing SEO or human visitor experience?

Not at all. Adding /.well-known/ endpoints, Link response headers, and robots.txt AI-crawler blocks are purely additive server-side changes invisible to browsers and human users. They carry no performance overhead and do not alter any existing URL structure, HTML, or Googlebot crawl behavior in any way.

What is the biggest opportunity if GitHub improves its AEO score?

GitHub hosts canonical source code and documentation for a vast share of open-source software. AI coding assistants and answer engines that can discover and cite GitHub content via a proper MCP Server Card or API catalog would surface repository READMEs, API references, and issue discussions in AI-generated answers far more frequently, reinforcing GitHub's authority as a primary developer knowledge source.