Unified Intent Mediator

A reference to a provider's Unified Intent Mediator (UIM) surface — an early-stage protocol proposal for standardizing how AI agents discover, negotiate, and execute intents against web services, rather than driving raw endpoints. UIM's contribution is that it treats the intent as the unit of interaction and pairs it with a policy layer covering permissions and compensation. It is a proposal, not an adopted standard — the specification is published openly under Apache-2.0 but has no ratifying body and no known production adopters, so record it as an experiment alongside Model Context Protocol rather than as a peer of it.

Also known as: UIM, UIM Protocol, Unified Intent Mediator Protocol

Example

- type: UnifiedIntentMediator
  url: https://developers.example.com/uim

Standards

Open proposal (Apache-2.0, no standards body) UIM Protocol specification
Open proposal (source of record) UIM Protocol repository
Anthropic / open specification Model Context Protocol
Linux Foundation Agent2Agent (A2A) Protocol
Community convention llms.txt
Community convention AGENTS.md
OpenAPI Initiative OpenAPI Specification

OpenAPI Expression

operationId (OpenAPI 3.x) — Stable operationIds are what any intent layer binds to; unstable ones break the mapping silently.
tags (OpenAPI 3.x) — Tag groupings are the usual starting point for deriving candidate intents from an existing API.
components.securitySchemes (OpenAPI 3.x) — The permission layer an intent protocol has to delegate to rather than replace.

Risk & Compliance

Security: Any intent-mediation layer moves authorization decisions away from the endpoint and toward a broker, which widens the blast radius of a misconfigured policy — an over-broad intent can compose several individually-safe operations into one unsafe action. UIM specifically proposes a policy and compensation layer, which means it also carries machine-negotiated commercial terms; treat those as an untrusted input path. Because the protocol is pre-adoption with no conformance suite, do not depend on it for access control. Keep enforcement at the API, scope agent credentials narrowly, and log intent execution against the underlying operations it resolved to.

Suggested Metrics

intents_declared — Count of intents a provider publishes.
intent_to_operation_coverage — Share of API operations reachable through a declared intent.
intent_resolution_success_rate — Fraction of agent intent requests that resolve to an executable operation.