Status: Research incubation. This repository does not contain a specification.
Nothing here is normative, stable, or ready for adoption.
This repository investigates one question:
How should digital services declare the terms under which automated reasoning systems may invoke them at runtime—and should those declarations become part of existing standards or require new standardization work?
The answer is intentionally unknown.
Evidence, rather than prior assumptions, determines the outcome.
Modern automated systems increasingly invoke external services at runtime:
- APIs
- MCP servers
- machine-readable tools
- transactional web interfaces
- other digital services and interfaces
From the provider's perspective, an important question emerges:
What should a service be able to declare about automated runtime invocation?
Examples include:
- whether automated invocation is allowed, restricted, or disallowed;
- how submitted inputs are handled;
- who is responsible for the service;
- how access may be revoked;
- what governance information should be available to consumers.
Whether these declarations belong in a new specification—or inside existing standards—is precisely what this repository investigates.
Current work focuses on:
- the provider-side perspective of automated service invocation;
- existing standards and protocols;
- identification of genuine gaps, if any;
- candidate vocabulary that could be contributed to existing ecosystems.
The repository deliberately starts from the assumption that no new standard should be created unless existing standards are demonstrably insufficient.
Research currently evaluates work including:
- IETF AIPREF
- TDM Reservation Protocol (TDMRep)
- W3C ODRL
- Model Context Protocol (MCP)
- Robots Exclusion Protocol (RFC 9309)
- AI-related robots extensions
- Agent2Agent (A2A)
- identity and credential frameworks
The objective is to understand what already exists before proposing anything new.
This research may ultimately produce:
- a proposal to an existing standard;
- a profile for an existing specification;
- a vocabulary contribution;
- recommendations only;
- or no new specification at all.
No outcome is predetermined.
Agent Manifest studies declarations made by autonomous actors.
This repository investigates the declaration needs of the services and interfaces consumed by those actors.
Whether those concerns deserve their own artifact—or belong inside existing standards—remains an open research question.
docs/provider-side-declaration-survey.md
Code is licensed under Apache-2.0 (see LICENSE).
Documentation and research material are licensed under CC BY 4.0 (see LICENSE-DOCS.md).
Hernán Alfredo Capucci
https://agent-manifest-spec.org
ORCID: 0009-0008-7216-3032