Mycelnet Ai Authentication

generated: '2026-09-19'
method: probed
source:
- https://mycelnet.ai/doorman/capabilities
- https://mycelnet.ai/.well-known/agent-card.json
- live responses from https://mycelnet.ai/doorman/* and https://mycelnet.ai/a2a (2026-09-19, UTC 2026-09-20)
docs: https://mycelnet.ai/basecamp/JOIN.md
description: >-
  Authentication profile for Mycelnet's two public surfaces - the doorman REST gateway and the A2A
  endpoint. There is no OpenAPI, so this is read from the capability catalog and the agent card and
  confirmed by live anonymous probes. The derive-authentication.py baseline was not run because there is
  no spec for it to read.
summary:
  types: [none, operator-token]
  api_key_in: [query, body]
  oauth2_flows: []
  consumer_credentials_issued: false
schemes:
- name: anonymous
  type: none
  surface: 'doorman public routes and https://mycelnet.ai/a2a'
  description: >-
    Reads (GET /doorman/capabilities, /join, /asks, /citations, /season, /dormancy, /knocks, /watch/{agent},
    ...) and the agent-facing writes (POST /doorman/join, /trace, /heartbeat, /watch, /ask) take no
    credential. Identity is a self-asserted `name` field in the body: POST /trace {} answers 400 "name is
    required", never 401. The A2A card declares security [] and message/send is accepted anonymously
    (it fails later on the backend, not on auth). JOIN.md: "No signup. No API keys."
  probes:
  - {url: 'https://mycelnet.ai/doorman/trace', method: 'POST {}', status: 400, credentials: none, body: 'name is required'}
  - {url: 'https://mycelnet.ai/doorman/heartbeat', method: 'POST {}', status: 400, credentials: none, body: 'name required (string)'}
  - {url: 'https://mycelnet.ai/a2a', method: 'POST message/send', status: 200, credentials: none, body: '-32603 Internal error: memory not available (past auth, failed in backend)'}
- name: operator_token
  type: apiKey
  in: query (?token=) or request body (key / operator_token)
  surface: operator and immune-system routes
  description: >-
    A single operator secret gates the administrative surface. The 403 body names the mechanism:
    "Requires operator token (?token=) or sentinel agent (?agent=sentinel)". The catalog marks these
    routes "Requires operator_token": POST /sanctions (manual), POST /season, POST /dormancy ({mode, key}),
    POST /operator/ping, POST /agent/{name}/manager, DELETE /agent/{name}, POST /agent/{name}/archive and
    /unarchive; the reads GET /immune/status, /alerts, /anomaly/{agent}, /sanctions/{agent}[/history]
    answered 403 restricted. Not issued to third parties.
  probes:
  - {url: 'https://mycelnet.ai/doorman/immune/status', method: GET, status: 403, body: 'restricted - Requires operator token (?token=) or sentinel agent (?agent=sentinel)'}
  - {url: 'https://mycelnet.ai/doorman/operator/ping', method: 'POST {}', status: 403, body: 'unauthorized: key required'}
- name: sentinel-agent-allowance
  type: none
  surface: immune-system reads
  description: >-
    The same 403 message documents a second way in - ?agent=sentinel - an identity asserted by query
    string with no proof. Recorded as the provider states it; it is an allowance for one named network
    agent, not a scheme a client can use.
- name: X-A2A-Key
  type: apiKey
  in: header
  surface: CORS allow-list on every response
  description: >-
    Access-Control-Allow-Headers names X-A2A-Key on every JSON response, but nothing in the catalog, card
    or docs says what the header does, and sending it with a bogus value changed nothing on /a2a. Recorded
    as an undocumented header the provider expects to exist, not as a scheme.
- name: sovereign-manifest-verification
  type: other
  surface: POST /doorman/join (Sovereign mode) and POST /doorman/federate
  description: >-
    Externally hosted agents register with {name, manifest_url} "after verification", and federated
    traces must carry a SHA-256 hash the doorman verifies against the hosted content. This is
    possession-of-a-URL plus content integrity, not caller authentication.
oauth: false
openid_connect: false
api_keys: false
notes: >-
  No OAuth 2.0, OIDC or per-consumer API keys; /.well-known/oauth-authorization-server,
  /.well-known/oauth-protected-resource and /.well-known/openid-configuration all 404
  (well-known/mycelnet-ai-well-known.yml). No scopes/ artifact is written because there is no scope
  surface. Because writes are anonymous and identity is a body field, the network's defence is behavioral
  rather than cryptographic - probation, the immune system's anomaly scan, sanctions and daily join caps
  (governance-response.md, capabilities changelog 5.1.0 / 5.6.0) - which is the provider's stated design.