| Internet-Draft | AgentManifest | July 2026 |
| Jernalczyk | Expires 7 January 2027 | [Page] |
AgentManifest defines a JSON document that websites can publish to describe identity, trusted knowledge, agent-facing capabilities, structured bindings, risk levels, consent requirements, authentication expectations, audit rules, and policies.¶
The goal is to help AI agents understand what a website knows and what it can safely do before scraping, guessing from visual UI, or executing brittle browser automation.¶
This note is to be removed before publishing as an RFC.¶
This document is an experimental individual Internet-Draft proposal for discussion. It is not an RFC and is not an IETF-approved standard.¶
Source material for this draft is maintained in the AgentManifest repository. The draft is expected to change based on implementation feedback and standards discussion.¶
This Internet-Draft is submitted in full conformance with the provisions of BCP 78 and BCP 79.¶
Internet-Drafts are working documents of the Internet Engineering Task Force (IETF). Note that other groups may also distribute working documents as Internet-Drafts. The list of current Internet-Drafts is at https://datatracker.ietf.org/drafts/current/.¶
Internet-Drafts are draft documents valid for a maximum of six months and may be updated, replaced, or obsoleted by other documents at any time. It is inappropriate to use Internet-Drafts as reference material or to cite them other than as "work in progress."¶
This Internet-Draft will expire on 7 January 2027.¶
Copyright (c) 2026 IETF Trust and the persons identified as the document authors. All rights reserved.¶
This document is subject to BCP 78 and the IETF Trust's Legal Provisions Relating to IETF Documents (https://trustee.ietf.org/license-info) in effect on the date of publication of this document. Please review these documents carefully, as they describe your rights and restrictions with respect to this document. Code Components extracted from this document must include Revised BSD License text as described in Section 4.e of the Trust Legal Provisions and are provided without warranty as described in the Revised BSD License.¶
AgentManifest is an experimental open specification. Implementers are encouraged to test the draft, report interoperability issues, and propose improvements.¶
The key words MUST, MUST NOT, REQUIRED, SHALL, SHALL NOT, SHOULD, SHOULD NOT, RECOMMENDED, MAY, and OPTIONAL in this document are to be interpreted as described in [RFC2119] and [RFC8174] when, and only when, they appear in all capitals, as shown here.¶
An AgentManifest document declares:¶
Capabilities describe what a website can do and under which policy.
Structured bindings describe how that capability can be accessed
or executed, such as through HTML, static resources, HTTP,
OpenAPI ,
MCP , or hosted checkout.
OpenAPI operations and MCP tools are binding targets, not the whole
capability contract.¶
AgentManifest is vendor-neutral. Implementation frameworks can generate manifests and bindings, but the manifest does not require any specific generator or expose implementation ownership as a normative field.¶
Agents SHOULD attempt static discovery at
/.well-known/agent.json. Publishers MAY also
expose /.well-known/agent-manifest.json,
/agent-manifest.json, or an HTML discovery hint:¶
<link rel="agent-manifest" href="/.well-known/agent.json">¶
If no manifest is found, agents SHOULD continue normal web behavior and SHOULD NOT infer that unsupported actions are safe.¶
The canonical JSON Schema for the repository version of this draft is:¶
rfc/schemas/agent-manifest.v0.1.schema.json¶
The schema uses JSON Schema Draft 2020-12
and keeps
additionalProperties: true for forward-compatible
experimentation.¶
A conforming manifest for this draft uses the
agentManifestVersion value 0.1-draft.¶
Capabilities declare a risk level of low, medium,
high, or critical.¶
Low-risk capabilities are public and read-only. Medium-risk capabilities submit data or start a workflow. High-risk capabilities modify user or business state. Critical capabilities include payments, orders, legal commitments, destructive actions, or irreversible operations.¶
High and critical actions MUST require explicit
confirmation and audit. Critical actions MUST set
stateChange to true, requiresConsent to
true, and consentMode to explicit,
step_up, or human_review.¶
AgentManifest declares authentication, authorization, consent, audit, and policy expectations. Protocol endpoints are responsible for enforcement. Consent must be specific to a capability invocation and must not be treated as broad permanent authorization. Critical actions should use appropriate authentication, explicit confirmation, audit, replay protection, and duplicate transaction controls.¶
Manifests and related public files MUST NOT expose secrets, private tokens, unpublished data, private customer information, admin-only endpoints, or privileged internal operations.¶
Servers MUST enforce authorization, consent, validation, replay protection, duplicate transaction controls, and policy checks outside prompt text. Prompt instructions are not a security boundary.¶
Publishers should minimize disclosed and collected data. Knowledge sources should include only information intended for public agent use. State-changing actions should collect only fields required for the declared capability.¶
Agents should provide user-visible summaries for submitted data and should avoid logging sensitive user information unless required for audit, compliance, or fraud prevention.¶
This document makes no IANA registration request.¶
The media type application/agent-manifest+json is proposed
for discussion only and is not registered by this document.¶