Skip to main content
v1.2Coming soon

Capability Graph: from a long tool list to the right next option

Private Beta available by request: request a demo.

When an API has dozens or hundreds of operations, showing every tool to an AI client is not always helpful. Capability Graph gives HAPI a structured view of what each operation is for, when it should be used, what it changes, and what it needs first.

HAPI can then offer a smaller, relevant set of tools for a request such as “move my appointment without cancelling it” or “show overdue invoices for this customer.”

Deterministic first, semantic when it helps​

The standard HAPI experience uses deterministic graph-backed discovery. It is predictable, reviewable, and respects authorization before a tool is suggested.

The v1.2 Private Beta also introduces an optional Enterprise classifier based on Jev from TypeSafe. Jev can help rank an already-authorized shortlist when the language is nuanced, ambiguous, or expressed in many different ways.

This is a deliberate split of responsibility:

  • The API contract says what is possible.
  • HAPI applies authorization and policy.
  • Capability Graph builds the eligible shortlist.
  • Jev helps understand intent and rank that shortlist.
  • The MCP client still decides how to continue the conversation.

Jev is not a replacement for permissions, confirmation, or business rules. Those remain explicit and deterministic.

What API teams provide​

Most APIs already have useful signals: tags, summaries, descriptions, inputs, responses, security schemes, and HTTP methods. HAPI can use those signals.

Teams can add a small amount of optional business context when it removes ambiguity:

  • what a tool helps a person accomplish;
  • when an assistant should use it or avoid it;
  • whether it reads, creates, updates, removes, or triggers an action;
  • which human-readable conditions should be true first.

This is not a rewrite of the backend. It is a clearer guide for the assistant using it.

Who benefits?​

Capability Graph is especially useful when:

  • an API has a large or growing MCP tool catalog;
  • several operations sound similar but have different consequences;
  • an assistant must avoid offering unauthorized tools;
  • a product team wants consistent tool discovery across MCP clients.

For the next layer—knowing whether the current situation is ready for an action—continue to Capability Planning.