The corpus, as a dataset.

Waymark's route map — how agents actually call real APIs, step by step, with the gotchas — is queryable free, forever. Licensing is for companies that want the dataset itself: in bulk, with redistribution and training rights, as an eval and training asset.

7,600+ documented routes 2,400+ API domains 140+ individually fact-checked Live API drift feed (JSON + RSS)

Counts as of July 8, 2026, and growing — live numbers at mcp.waymark.network/stats.

What's in a route record

Every record is a task-level procedure for a real API: the ordered steps that work today, the documented failure modes, and machine-readable trust metadata. Sample shape (full schema in the OpenAPI spec):

{
  "task": "verify a Stripe webhook signature",
  "domain": "stripe.com",
  "steps": [ "…ordered, current, actionable…" ],
  "gotchas": [ "…the failure modes that burn agents…" ],
  "verification": { "status": "verified | sampled | unverified" },
  "attestations": { "success": 13, "failure": 0, "keyed_success": 2 },
  "contributor": "…", "created": "…"
}

Plus the API drift feed: a growing log of observed API behavior changes that silently break agents running on stale knowledge, each with the re-verified fix, in JSON and RSS.

Why license what's free to query?

Bulk, versioned snapshots

The public API serves live queries, one task at a time. A license gets you the full corpus as versioned snapshot exports on your schedule, with a provenance manifest — every record's verification status, attestation history, and contribution timestamps.

Redistribution & training rights

Free public access doesn't include redistributing the corpus or embedding it into a commercial product or model. A license grants the rights your legal team will ask about: internal use, product embedding, eval-set redistribution, or model training — scoped to what you need.

An eval asset, not just data

Routes pair naturally into task → expected-procedure eval sets for tool-use and agentic benchmarks. Our own blind-graded benchmark methodology is published; the retrieval eval harness ships in the open-source repo.

Who this is for

Agent-framework companies

  • Ship a procedural-knowledge layer without building a corpus from zero
  • Benchmark your agents' tool-use against documented real-API procedures

Model labs

  • Training and eval data for API-procedural knowledge — current, structured, provenance-tagged
  • Drift feed as a freshness signal: which API knowledge went stale, when, and what changed

Devtools & API observability

  • Per-domain procedure coverage across 2,400+ API domains
  • Drift events as an early-warning signal for breaking API changes

What we won't oversell

Provenance is per-record and honest: 140+ routes are individually fact-checked against current provider docs; the rest carry sampled or unverified status, and every record states which. Attestation coverage is early. The drift feed is young. You'll see exactly what you're licensing before you sign — the whole corpus is publicly inspectable at mcp.waymark.network/routes.

Annual license, scoped to your use

Pricing depends on scope (internal eval use vs. product embedding vs. training rights) and delivery cadence. One email starts the conversation — you'll talk to the operator, not a sales sequence.

Email [email protected]