# agents.txt

> A root-level file that declares a site's identity, terms of use for AI agents, service catalog and agentic endpoints in a machine-readable form.

_The Agent Protocol Atlas · /protocols/agents-txt · [JSON](/api/protocols/agents-txt) · [all The Agent Protocol Atlas](/protocols)_

- **name:** agents.txt
- **full_name:** agents.txt
- **layer:** discovery
- **creator:** open community (multiple competing proposals; asturwebs reference among them)
- **status:** draft (community proposal)
- **year:** 2025
- **one_liner:** A root-level file that declares a site's identity, terms of use for AI agents, service catalog and agentic endpoints in a machine-readable form.
- **spec_url:** https://github.com/asturwebs/agents-txt
- **snippet:**

```
GET /agents.txt → identity, permissions, services, agentic endpoints (with /api/agents JSON twin)
```
- **abbreviation:** agents.txt
- **also_known_as:** /agents.txt, agent-manifest.txt (proposed 2026 rename of one variant)
- **canonical_spec_url:** https://github.com/asturwebs/agents-txt
- **entity_uri:** https://github.com/asturwebs/agents-txt
- **taxonomy_layer:** discovery
- **sub_layer:** agent-capability-declaration
- **protocol_type:** declaration
- **central_problem:** Gives a site a sanctioned, machine-readable channel to tell AI agents who it is, what they are permitted to do, and which agentic endpoints exist — instead of agents guessing.
- **maintainer:** Community proposal (asturwebs reference implementation; other variants by kaylacar, dennj, and the agent-manifest.txt rename)
- **governance_body:** none (community proposal; IANA well-known registration pending review)
- **license:** MIT
- **maturity_tag:** experimental
- **current_spec_version:** 2.0 (asturwebs variant, draft)
- **spec_date:** — (verify-against-primary-at-build)
- **launch_date:** 2025
- **last_verified:** 2026-06-15
- **transport:** Static file at the domain root (/agents.txt) with a JSON twin (/api/agents)
- **core_mechanism:** A site serves a human-readable, machine-parseable file at /agents.txt declaring its identity, AI-agent terms of use, service catalog, and agentic endpoints (with a JSON twin); agents fetch it to learn what they are permitted to do. Note: multiple competing agents.txt proposals exist and one variant was renamed agent-manifest.txt in early 2026, so the format is not yet converged.
- **discovery_endpoint:** /agents.txt (and /.well-known/agents.txt per pending IANA registration)
- **settlement_type:** —
- **adoption_metric:** — (verify-against-primary-at-build)
- **notable_adopters:** {"value":"AsturWebs (reference implementation)","source":"https://github.com/asturwebs/agents-txt"}
- **relationships:** {"predicate":"discovers","target":"mcp","note":"agents.txt -discovers-> MCP (research §2 seed triple): an agents.txt-class discovery file can point agents to a site's MCP / agentic endpoints."}, {"predicate":"complements","target":"llms-txt","note":"agents.txt declares agent permissions/endpoints; llms.txt declares readable content — paired Layer-1 discovery files."}
- **ideal_use_case:** Declaring, in one root file, what AI agents are allowed to do on your site and where your agentic endpoints live.
- **when_to_use:** When you want to publish agent permissions and a service/endpoint catalog beyond what llms.txt (content) or robots.txt (crawl rules) cover.
- **when_not_to_use:** When you only need to expose readable content (use llms.txt) or you require a converged, stable standard — agents.txt has multiple competing drafts.
- **code_example:** # /agents.txt
Name: Acme
Policy: agents may read and call the catalog API; no scraping of /account
Services:
  - search: /api/search
  - checkout: /.well-known/agents.json#checkout
- **source:** Reference variant (MIT, v2.0 draft, identity/permissions/services/endpoints, /agents.txt + /api/agents, IANA well-known pending, agent-manifest.txt rename of one variant): https://github.com/asturwebs/agents-txt . Seed triple (agents.txt discovers MCP): research §2. Honesty note: multiple competing agents.txt proposals exist.
- **agent_readiness_link:** agent-readiness/discoverability
- **layer_legacy:** content