Internet-Draft BotCentral Card September 2026
Mitchell Expires 5 March 2027 [Page]
Workgroup:
Network Working Group
Internet-Draft:
draft-mitchell-botcentral-card-00
Published:
Intended Status:
Informational
Expires:
Author:
R. Mitchell
BotCentral

The BotCentral Card: An Owner-Proven Consent Record for Automated Web Clients

Abstract

The Robots Exclusion Protocol (RFC 9309) lets a site say which automated clients may fetch its content. It cannot express purpose: fetching a page to answer a person is not the same act as copying it into a model training set or taking an action on the site. It also cannot prove who published the policy.

This document defines the BotCentral Card, a JSON record, one per domain, that states which purposes the domain owner consents to ("retrieve", "train", and "act" as separate answers), backed by a proof of domain control placed either in a DNS TXT record or at the well-known URI "/.well-known/botcentral.txt". Cards are written to a registry by authenticated publishers and read by any client over HTTP or the Model Context Protocol. Clients never write cards. A card is permission to be found; it is not a ranking and not a training grant.

This document also registers the "botcentral.txt" well-known URI.

Status of This Memo

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 5 March 2027.

Table of Contents

1. Introduction

Search engines and web sites share a discovery layer that has accumulated over two decades: the Robots Exclusion Protocol [RFC9309], sitemaps, IndexNow [INDEXNOW], and, in advertising, "ads.txt" [ADSTXT]. Each of these is a small, public, owner-placed file that a large population of clients agrees to read.

Automated clients that fetch the web on behalf of AI assistants have no such layer. They have "robots.txt", which was designed for crawlers and answers one question: may this product token fetch this path? It cannot say "you may retrieve this page to answer a user, but you may not train on it, and you may not act on it". It cannot say who owns the site. And it cannot be queried across the web; it must be fetched one origin at a time.

In the absence of a shared record, every operator of such clients has built a private one. Each ships its own product tokens, often several per vendor to distinguish training from retrieval from search, together with its own submission form and its own allowlist. Site owners face an unbounded number of portals. Client operators cannot keep an unbounded number of allowlists honest.

This document defines a single record, the BotCentral Card, that answers those questions once. The record is built from mechanisms that already exist: "robots.txt" [RFC9309] for crawl permission, DNS TXT records and well-known URIs [RFC8615] for proof of control, the TDM Reservation Protocol [TDMREP] for training reservation, OAuth 2.0 client credentials [RFC6749] for publisher authentication, and HTTP entity tags [RFC9110] for caching. The card adds the one thing those mechanisms lacked: a queryable index of who opted in, for which purpose, with proof.

The design deliberately follows the shape of "ads.txt" and of the Web PKI: a proof on the origin, a public registry, and clients that refuse unsigned claims.

1.1. Requirements Language

The key words "MUST", "MUST NOT", "REQUIRED", "SHALL", "SHALL NOT", "SHOULD", "SHOULD NOT", "RECOMMENDED", "NOT RECOMMENDED", "MAY", and "OPTIONAL" in this document are to be interpreted as described in BCP 14 [RFC2119] [RFC8174] when, and only when, they appear in all capitals, as shown here.

1.2. Terminology

Owner
The party that controls a domain, demonstrated by the proof in Section 4.
Publisher
A party authorized by an owner to write cards to a registry. Publishers authenticate with OAuth 2.0 client credentials [RFC6749]. A publisher is analogous to a seller in "ads.txt".
Registry
A service that verifies proof, stores cards, and serves them. The reference registry is operated at https://botcentral.org.
Client
Any program that reads cards: a crawler, an assistant, a Model Context Protocol [MCP] host, or a search engine.
Card
The JSON record defined in Section 3.

2. Roles and Trust Model

Owners prove control of a domain. Publishers write cards on behalf of owners. Registries verify proof, apply defaults, compute derived fields, and serve cards. Clients read cards and honor them.

A registry MUST NOT accept a card write from a client. Only publishers write. A card written by a publisher does not become proven by being written; it becomes proven only when the registry independently confirms the proof in Section 4.

3. The Card

A card is a JSON object served with media type "application/json". A registry MUST publish the card's JSON Schema [JSONSCHEMA] at "/v1/schema.json". The normative field list is in [BOTCENTRAL]; this section summarizes the members whose semantics matter for interoperability.

3.1. Required Members

botcentral
The card version, "1.0" or "1.1".
domain
The registrable domain in lower case.
canonical
The site's canonical origin. MUST use the "https" scheme.
name, summary, topics
Display name, one-paragraph description, and free-form topic tags.
allow_bots
Boolean master switch. When false, the registry MUST serve "retrieve", "train", and "act" as false regardless of what was submitted.
allow, deny
Lists of "robots.txt" product tokens the owner explicitly allows or denies.
pointers
Absolute URIs of the site's own "robots.txt", sitemap, and "llms.txt". Pointers are informational and are not proof.
pages
Declared URLs, each with a "url" and a "rel" role, and optionally a registry-computed "score" (Section 3.3).
updated
An [RFC3339] timestamp of the last publisher write.
verification
The proof state (Section 4).
consent
The purpose consent block (Section 3.2).

3.3. Scores

Each declared page MAY carry a "score", an integer from 0 to 100 computed by the registry from proof, consent, freshness, crawl health, page role, and surface. It is a hint for crawl ordering. Publishers MUST NOT set it, registries MUST ignore a submitted value, and clients MUST NOT present it as a ranking of sites against one another or as a claim about content quality.

3.4. Extensibility

Unknown members MUST be preserved by registries and ignored by clients.

4. Proof of Control

A card is proven by exactly one of two methods. Nothing else counts.

4.1. Verification Token

The registry issues the publisher a per-domain verification token: an opaque string carrying at least 128 bits of entropy, encoded in URL-safe characters, bound to the pair (registry, domain).

4.2. Method "well-known-file"

The owner serves the token at the well-known URI:

https://<domain>/.well-known/botcentral.txt

All of the following MUST hold for the proof to pass:

  1. The HTTP status is 200.
  2. The response "Content-Type" is not "text/html".
  3. The first 400 octets of the body, compared case-insensitively, do not contain "<!doctype", "<html", "<head", or "<body".
  4. The body contains the line "botcentral-verify=<token>".

Conditions 2 and 3 exist because single-page-application hosts routinely return status 200 with an HTML shell for every path. An HTML 200 is a failed proof, not a passed one.

The file MAY contain multiple "botcentral-verify=" lines, one per registry, and MAY contain comment lines beginning with "#".

4.3. Method "dns-txt"

The owner publishes a DNS TXT record at the apex of the domain whose data contains "botcentral-verify=<token>". Multiple TXT records at the apex are normal; the registry MUST scan all of them. Registries SHOULD resolve through a validating resolver. The choice of resolver is outside the scope of this document.

4.4. Precedence and States

A registry checks Section 4.2 first, then Section 4.3. The first success is recorded in "verification.method". If neither succeeds, "verification.method" is one of:

pending
A token has been issued and no check has completed.
unverified
A check completed and failed. "verification.note" states why: origin unreachable, "robots.txt" unreachable, token absent, HTML shell returned, or "robots.txt" contains "Disallow: /" for a product token the card lists in "allow".
stale
A prior proof has not been re-confirmed within the freshness window (Section 5).

"verification.checked" carries the [RFC3339] time of the last check, or null.

4.5. What Is Not Proof

Submission through a publisher, including a registry operator's own publisher, is not proof. A reachable origin is not proof. A sitemap or "llms.txt" is not proof. An HTML "meta" element is not proof, deliberately: on a shared host any tenant can emit one, whereas the two accepted methods require control of DNS or of the "/.well-known/" path, which shared tenants lack.

A registry MUST NOT serve "verification.method" of "dns-txt" or "well-known-file" for any card that did not pass Section 4.3 or Section 4.2 respectively.

5. Freshness

Proof decays. A registry re-checks proof periodically. The default window is 168 hours; an owner MAY set "crawl.revalidate_hours" to another value.

fresh
Proven, and the last check is within the window.
stale
Proven previously; the last check is older than the window.
unproven
"verification.method" is "pending" or "unverified".

Clients SHOULD treat "stale" cards as listed but of lower confidence, and SHOULD treat "unproven" cards as unlisted for any purpose that depends on owner consent.

6. Registry Interface

A registry exposes the following, relative to its origin. Anonymous reads MUST be permitted, subject to rate limiting.

6.1. Discovery

GET /.well-known/botcentral-index.json

Returns an object with the card version, the registry name, and absolute URIs for the catalog search, the contract documentation, the OpenAPI description, the schema, the change feed, the snapshot, and the MCP endpoint, together with a "listed" array of {domain, name, href} entries. All other registry resources are discoverable from this document.

6.2. Read

GET /v1/site/{domain}
GET /v1/search?q=&topic=&verified=&consent=&capability=
GET /v1/changes
GET /v1/snapshot
GET /v1/schema.json
GET /v1/openapi.json

"/v1/site/{domain}" returns one card and MUST support "ETag" and "If-None-Match" [RFC9110]. "/v1/changes" is a feed of (domain, updated, etag) tuples since a cursor. "/v1/snapshot" returns a hash over the whole catalog so that a client can ask "has anything changed?" in one request, in the spirit of IndexNow [INDEXNOW].

6.3. Model Context Protocol

A registry SHOULD expose an MCP [MCP] server at "POST /mcp", advertised at "/.well-known/mcp.json", offering at minimum the tools "search_catalog" and "get_listed_site" with semantics identical to the read interface.

6.4. Write

POST /oauth/token
PUT  /v1/publisher/sites/{domain}
POST /v1/publisher/sites/{domain}/reverify

Publishers obtain tokens with the OAuth 2.0 client credentials grant [RFC6749]; authorization server metadata is published per [RFC8414] and protected resource metadata per [RFC9728]. A write MUST NOT make a card fresh; only a passed proof check does that.

6.5. Compiled robots.txt

A registry MAY compile a "robots.txt" [RFC9309] from a card for the owner's convenience. When it does, product tokens whose purpose is training receive "Disallow: /" unless "consent.train" is true; tokens whose purpose is retrieval remain allowed unless "consent.retrieve" is false; entries in "allow" and "deny" override. Some tokens, such as "Google-Extended" and "Applebot-Extended", are control tokens that never appear as HTTP User-Agent values; a registry that publishes an agent list MUST label them as such.

7. Client Obligations

A client that reads a card and then fetches the origin:

  1. MUST honor "consent" for its actual purpose. A retrieval fetch is permitted by "retrieve" true; using the result for training requires "train" true.
  2. MUST still honor the origin's own "robots.txt". The card never overrides it.
  3. SHOULD send "If-None-Match" on repeat card fetches and poll "/v1/snapshot" rather than re-read the catalog.
  4. MUST NOT treat "score" as a ranking or a quality claim.
  5. SHOULD attribute the origin when "cite" is true.

8. Security Considerations

8.1. Server-Side Request Forgery

A registry fetches owner-supplied URLs to verify proof. It MUST resolve the target hostname before connecting and reject private, loopback, link-local, and cloud metadata address ranges; MUST require the "https" scheme; MUST cap redirects; and MUST enforce a timeout and a response body size limit.

8.2. HTML Shells

Single-page-application hosts return status 200 for every path. The HTML rejection rules in Section 4.2 are normative for this reason. A registry that accepted an HTML 200 as proof would let any tenant of such a host claim any path.

8.3. Shared Hosting

Page content and "meta" elements are excluded from proof because any tenant of a shared host can emit them. The two accepted methods require control of DNS or of the "/.well-known/" path.

8.4. Token Replay and Domain Transfer

A token proves control at check time only. The freshness window in Section 5 bounds how long a lapsed or transferred domain keeps a proven card. Registries SHOULD shorten the window for domains whose registration data changes.

8.5. Publisher Compromise

A publisher can rewrite the cards it controls. Registries MUST record the writing publisher in "listed_by", MUST log writes, and SHOULD provide owners a means to revoke a publisher's authority.

8.7. Score Gaming

"score" is derived from proof, consent, freshness, crawl health, page role, and surface, none of which a publisher sets directly. Registries MUST NOT let a publisher influence "score" except through those underlying facts.

8.8. Privacy

Cards contain only information the owner chose to publish about a domain. A registry's change feed reveals when a domain's policy changed, which is intended. Registries SHOULD NOT log the identity of anonymous readers beyond what rate limiting requires.

9. IANA Considerations

9.1. Well-Known URI Registration

IANA is requested to register the following in the "Well-Known URIs" registry established by [RFC8615]:

URI suffix
botcentral.txt
Change controller
Rodney Mitchell (rodneymitchellv@gmail.com)
Specification document
This document, Section 4.2
Status
permanent
Related information
The resource is plain text and MUST NOT be served with media type "text/html". See Section 4.2.

9.2. Well-Known URI Registration: botcentral-index.json

IANA is requested to register the following in the same registry:

URI suffix
botcentral-index.json
Change controller
Rodney Mitchell (rodneymitchellv@gmail.com)
Specification document
This document, Section 6
Status
permanent

10. References

10.1. Normative References

[JSONSCHEMA]
Wright, A., Andrews, H., Hutton, B., and G. Dennis, "JSON Schema: A Media Type for Describing JSON Documents", , <https://json-schema.org/draft/2020-12/json-schema-core>.
[RFC2119]
Bradner, S., "Key words for use in RFCs to Indicate Requirement Levels", BCP 14, RFC 2119, DOI 10.17487/RFC2119, , <https://www.rfc-editor.org/info/rfc2119>.
[RFC8174]
Leiba, B., "Ambiguity of Uppercase vs Lowercase in RFC 2119 Key Words", BCP 14, RFC 8174, DOI 10.17487/RFC8174, , <https://www.rfc-editor.org/info/rfc8174>.
[RFC8414]
Jones, M., Sakimura, N., and J. Bradley, "OAuth 2.0 Authorization Server Metadata", RFC 8414, DOI 10.17487/RFC8414, , <https://www.rfc-editor.org/info/rfc8414>.
[RFC9728]
Jones, M.B., Hunt, P., and A. Parecki, "OAuth 2.0 Protected Resource Metadata", RFC 9728, DOI 10.17487/RFC9728, , <https://www.rfc-editor.org/info/rfc9728>.

10.2. Informative References

[ADSTXT]
IAB Technology Laboratory, "ads.txt Specification Version 1.1", , <https://iabtechlab.com/ads-txt/>.
[BOTCENTRAL]
Mitchell, R., "BotCentral Card Specification 1.1", , <https://github.com/mitchvac/botcentral/blob/main/SPEC.md>.
[INDEXNOW]
Microsoft Bing and Yandex, "IndexNow Protocol", , <https://www.indexnow.org/documentation>.
[MCP]
Anthropic, PBC, "Model Context Protocol Specification", , <https://modelcontextprotocol.io/specification>.
[RFC3339]
Klyne, G. and C. Newman, "Date and Time on the Internet: Timestamps", RFC 3339, DOI 10.17487/RFC3339, , <https://www.rfc-editor.org/info/rfc3339>.
[RFC6749]
Hardt, D., Ed., "The OAuth 2.0 Authorization Framework", RFC 6749, DOI 10.17487/RFC6749, , <https://www.rfc-editor.org/info/rfc6749>.
[RFC8615]
Nottingham, M., "Well-Known Uniform Resource Identifiers (URIs)", RFC 8615, DOI 10.17487/RFC8615, , <https://www.rfc-editor.org/info/rfc8615>.
[RFC9110]
Fielding, R., Ed., Nottingham, M., Ed., and J. Reschke, Ed., "HTTP Semantics", STD 97, RFC 9110, DOI 10.17487/RFC9110, , <https://www.rfc-editor.org/info/rfc9110>.
[RFC9309]
Koster, M., Illyes, G., Zeller, H., and L. Sassman, "Robots Exclusion Protocol", RFC 9309, DOI 10.17487/RFC9309, , <https://www.rfc-editor.org/info/rfc9309>.
[TDMREP]
W3C TDM Reservation Protocol Community Group, "TDM Reservation Protocol (TDMRep)", , <https://www.w3.org/community/reports/tdmrep/CG-FINAL-tdmrep-20240202/>.

Appendix A. Example Card

{
  "botcentral": "1.1",
  "domain": "fieldnote.press",
  "canonical": "https://fieldnote.press/",
  "name": "Field Note",
  "summary": "Independent reporting on local government.",
  "topics": ["news", "infrastructure"],
  "allow_bots": true,
  "allow": ["ClaudeBot", "OAI-SearchBot"],
  "deny": [],
  "pointers": {
    "robots": "https://fieldnote.press/robots.txt",
    "sitemap": "https://fieldnote.press/sitemap.xml",
    "llms": "https://fieldnote.press/llms.txt"
  },
  "pages": [
    { "url": "https://fieldnote.press/", "rel": "home", "score": 94 }
  ],
  "updated": "2026-08-30T14:02:11Z",
  "verification": {
    "method": "well-known-file",
    "checked": "2026-08-31T09:10:00Z",
    "note": "Token matched /.well-known/botcentral.txt (plain text)."
  },
  "consent": {
    "retrieve": true,
    "train": false,
    "act": false,
    "cite": true,
    "tdm": "reserved"
  },
  "freshness": {
    "status": "fresh",
    "checked": "2026-08-31T09:10:00Z",
    "revalidate_hours": 168
  },
  "listed_by": { "id": "citefleet", "name": "CiteFleet" }
}

Appendix B. Example Proof Files

A "/.well-known/botcentral.txt":

# BotCentral proof of control. Plain text. Not HTML.
botcentral-verify=Kx7mQ2vB9pR4tY8wZ1aC3dF6gH0jL5nP

A DNS TXT record at the apex:

fieldnote.press. 3600 IN TXT
    "botcentral-verify=Kx7mQ2vB9pR4tY8wZ1aC3dF6gH0jL5nP"

Appendix C. Relationship to Existing Mechanisms

robots.txt [RFC9309]
Answers "may this token fetch this path?". The card answers "for what purpose, and says who?". The card never overrides "robots.txt".
TDM Reservation Protocol [TDMREP]
Expresses a training reservation for an origin. The card carries the same value in "consent.tdm" and defers to the origin's own file on conflict.
ads.txt [ADSTXT]
The model for this design: an owner-placed plain-text file on the origin, authorized sellers, and buyers that refuse unlisted claims.
IndexNow [INDEXNOW]
The model for the change feed and snapshot: tell clients what changed instead of making them recrawl.

Appendix D. Acknowledgements

The structure of this document follows the guidance at authors.ietf.org. The "ads.txt" and IndexNow designs informed the trust model and the change feed respectively.

Author's Address

Rodney Mitchell
BotCentral