Building an Agent-Native Registry: 309 Entries, Zero Runtime Dependencies

A machine-readable media property engineered for AI search — llms.txt, OpenAPI, MCP and JSON-LD, verified by 2,793 automated checks

6 min read
Published September 8, 2026
generative engine optimisationGEOllms.txtMCP serverJSON-LDAI searchagent-native

TL;DR

The GTM Stacker Agent Registry (browse it here) is a registry of 309 open-source go-to-market and RevOps tools, designed to be read by AI agents as easily as by people: llms.txt discovery files, OpenAPI 3.1 description, an MCP server, per-entry JSON-LD, sharded sitemaps, IndexNow submission and delta-sync feeds — all generated by a zero-runtime-dependency, deterministic static build and verified by 2,793 automated consumer checks plus a unit-test suite.

The thesis

AI assistants are becoming the front door to tool discovery. Most websites fight this; an agent-native property embraces it — publishing every entry in HTML for people and in Markdown, JSON and JSON-LD for machines, with stable identifiers and explicit provenance so an AI engine can cite it with confidence.

Architecture

  • Static generation with zero runtime dependencies — the whole build is auditable and rebuilds byte-identically.
  • Discovery surfaces: llms.txt and llms-full.txt, RFC 9727 api-catalog, OpenAPI 3.1, Agent Skills, and an MCP server exposing search and per-entry resources.
  • Structured data: JSON-LD SoftwareApplication per entry plus an Organization/founder entity graph.
  • Freshness machinery: sharded sitemaps with lastmod, IndexNow submission, conditional GET, and by-date delta-sync so agents fetch only what changed.
  • Governance: a schema that forces "unknown" over guesses, an explicit caveats field, and a lint pass against fabricated content.

Verification

Every build runs a consumer-simulation harness — 2,793 checks covering cold-start discovery, JSON parseability, link integrity and feed conformance — plus unit tests over the schema and content pipeline. A bad entry cannot ship.

Honest status

The architecture is complete and verified; curation is the ongoing work. A core of hand-written entries is being expanded across the bulk-imported catalogue, and the distribution plan (directories, search-console verification, syndication) is rolling out. This page will be updated as the registry accumulates citation and traffic data — claims will follow evidence, not precede it.

Browse it at gtmstacker.com/registry/ — or point your MCP client at it.
Share:
About the Author

Theo Popov runs GTM Stacker, a growth-systems practice for startups and venture programs. Previously Co-Founder & Operations Lead of an 11-location, ~$4M hospitality group bootstrapped with no outside capital, he now builds AI-native enrichment, outbound, content and reporting systems. More about Theo →

← Back to Case Studies