Toward Composition-Aware Operational Assurance for MCP: Observable Control Gaps, Outcome-Evidence Classes, and a Status-Bounded Design-Science Case

Abstract

Composable Model Context Protocol (MCP) workflows require assurance beyond endpoint reachability. This position and design-science preprint defines observable objects and bounded closure conditions for three related problems: the Registry-Runtime Gap, pre-execution and runtime Composition-Control Gaps, and the Execution-Outcome Gap between provider-reported success and evidence of the intended target state. The framework separates controls required by a plan, controls evidenced as configured before execution, and controls observed during or after an execution trace. It replaces an ordinal verification-tier model with non-ordered Outcome Evidence Source Classes evaluated across source authority, organizational independence, technical failure independence, cryptographic integrity, temporal proximity, identity binding, and observation completeness. A fully synthetic support-notice workflow demonstrates how missing task cancellation, provider substitution, and contradictory delivery evidence leave different gaps open. The accompanying public design supplement provides a generic Operational Assurance Profile JSON Schema, a synthetic example, bounded structural validation, and tests. SaSame is included only as an author-reported, status-bounded design case frozen at an exact internal branch snapshot. Observation and evidence components existed on main, while plan contracts, provider observation, dynamic resolution, composition planning, deterministic adapters, and a read-only execution lab existed only in an unmerged, non-production prototype. The paper does not claim that SaSame reduces risk, safely composes tools in production, or has established external-customer validity. Research status: Preprint v0.3.0. This work has not been peer reviewed. The proposed profile has not been validated for policy utility or inter-rater reliability, and no controlled comparative experiment currently establishes production effectiveness. Data and code availability: The accompanying supplement contains only a generic schema, a fully synthetic example, standard-library validation code, tests, citation metadata, and publication-boundary notices. It excludes private SaSame implementation, endpoint or observation data, provider-selection and composition logic, production security controls, credentials, customer information, internal infrastructure, pricing, entitlements, and commercial fulfillment systems. Conflict of interest: The author is affiliated with SaSame S.R.L., which designs and operates the case-study systems and offers commercial MCP services. The SaSame case is not an independent evaluation. Funding: This work received no external research funding.