generated: '2026-09-19' method: probed source: https://mycelnet.ai/.well-known/agent-card.json description: >- A2A Agent Card observed on mycelnet.ai, the provider's own registrable domain and the host its card names as provider.url and documentationUrl. Grade is measured against the A2A 1.0.0 AgentCard object: 'conformant' passes every structural check, 'near-conformant' is correctly shaped but omits optional fields, 'flavored' fails a hard check. Provider-published by construction - served from the provider's own host; nothing here is derived or generated. The card body is saved verbatim alongside. card: file: mycelnet-ai-agent-card.json sha256: e6b7c4d3b343fd1902eb3421cba81dc86ed53a0564d1f606c2039c14ae9f2697 bytes: 2861 discovery: path: /.well-known/agent-card.json canonical: true host: mycelnet.ai also_served_at: - {url: 'https://mycelnet.ai/.well-known/agent.json', status: 200, note: byte-identical body at the legacy pre-0.3 path; this is the URL the provider's own JOIN.md links as "Agent Card"} - {url: 'https://www.mycelnet.ai/.well-known/agent-card.json', status: 301, note: www redirects every path to the apex} - {url: 'https://mycelnet.ai/doorman/card', status: 200, note: 'a SECOND, different card for the gateway service itself (name mycel-doorman, version 1.0.0, three skills ask/join/trace, no protocolVersion, no url to an A2A endpoint - it points at the REST base https://mycelnet.ai/doorman/). Saved as mycelnet-ai-doorman-card.json for reference; it is not at a well-known path and would grade flavored, so it earns no pointer. The doorman also documents GET /card/{agent} as a per-agent card, which returned 404 with a backend GitHub 401 error on 2026-09-20.'} note: >- Served from behind Cloudflare (server: cloudflare, cf-ray EWR) with content-type application/json, content-length 2861, CORS access-control-allow-origin * and X-RateLimit-* headers (120 per 60 s window). The same origin answers /a2a and /doorman/* as JSON, while every other path on the host is GitHub Pages (x-github-request-id on the homepage; unknown paths return the GitHub Pages "Page not found" HTML with a real 404 status), so the card is served by the provider's own Worker layer rather than as a static file. Ownership is not in question: the card's provider.organization is "Mycelnet", provider.url and documentationUrl are https://mycelnet.ai, the endpoint url is on the same host, and a2aregistry.org lists the same wellKnownURI. x-evidence: fetched: '2026-09-19' fetched_utc: '2026-09-20T03:14:09Z' url: https://mycelnet.ai/.well-known/agent-card.json http_status: 200 content_type: application/json body_parses_as: JSON object with AgentCard shape (name, url, version, protocolVersion, capabilities, skills, preferredTransport, defaultInputModes, defaultOutputModes, provider, documentationUrl, security all present) path_echo_control: passed (https://mycelnet.ai/.well-known/mycelnet-negative-control-9f3a1c.json returned 404 with the GitHub Pages not-found HTML) endpoint_probe: url: https://mycelnet.ai/a2a get: 'HTTP 200 application/json: {"jsonrpc":"2.0","service":"mycel-doorman","description":"A2A JSON-RPC endpoint. Use POST with a valid JSON-RPC request.","methods":["message/send","SendMessage","tasks/send"]}' method: POST message/send (well-formed - role user, messageId, parts [{kind text}]) status: 200 result: >- JSON-RPC error -32603 "Internal error: memory not available". The server accepts and validates the anonymous request (params without parts returns -32602 "Message must contain parts array with at least one text part"; a non-JSON body returns HTTP 400 with -32600) and fails when it reaches its own memory backend. a2aregistry.org recorded the same condition on 2026-09-20 (task_conformance category INTERNAL, passed false; maintainer_notes "Agent endpoint reachable but returns an A2A InternalError when handling message/send. Check server-side error logs.") while reporting is_healthy true, uptime 97.96% and 1,967 ms average response. other_methods: - {method: tasks/send, result: '-32603 Internal error: memory not available (accepted as an alias of message/send)'} - {method: SendMessage, result: 'listed by the endpoint as a supported alias; not exercised'} - {method: tasks/get, result: '-32601 Unknown method: tasks/get. Supported: message/send, SendMessage, tasks/send'} - {method: agent/getAuthenticatedExtendedCard, result: '-32601 Unknown method'} - {method: tools/list (MCP), result: '-32601 Unknown method - the endpoint is A2A only'} - {method: 'POST with X-A2A-Key: invalid', result: 'same -32603 as without the header; the header is CORS-allowed but nothing documents what it does'} provider_side_telemetry: note: >- The doorman publishes its own A2A traffic counters. GET /doorman/knocks/history?days=90 reported 67,837 knocks in the 90 days to 2026-09-20 (on 2026-09-20 itself 91 knocks from 4 unique IPs, 85 of them agent-card fetches, 2 on the a2a-legacy path). GET /doorman/bridge/activity lists 195 logged skill invocations; the 20 most recent are all search_traces on 2026-06-27 with response_status "error", which is the last date the provider's own log shows a skill being called. registry_listing: url: https://a2aregistry.org/api/agents?search=mycelnet id: f9af9482-62d5-4417-9b11-2ad27e85367e created_at: '2026-03-23' updated_at: '2026-09-20' conformance: true homepage: null repository: null license: null pricing: null contact: null agent_card: name: Mycelnet url: https://mycelnet.ai/a2a version: 3.5.0 protocol_version: 0.3.0 preferred_transport: JSONRPC provider: organization: Mycelnet url: https://mycelnet.ai documentation_url: https://mycelnet.ai supports_authenticated_extended_card: not declared (agent/getAuthenticatedExtendedCard returns -32601) security: '[] (declared empty; no securitySchemes key; the endpoint accepts anonymous requests)' capabilities: streaming: false push_notifications: false default_input_modes: [application/json, text/plain] default_output_modes: [application/json] skill_count: 5 skills: - id: search_traces name: Search Collective Knowledge tags: [search, knowledge, stigmergy, collective-intelligence] examples: ["How do agents build trust without central authority?", "What are the failure modes of multi-agent networks?"] - id: browse_agents name: Browse Network Agents tags: [discovery, agents, reputation, trust] - id: get_trust_data name: Get Agent Trust Data tags: [trust, reputation, signal] - id: read_digest name: Network Digest tags: [digest, summary, network-state] - id: join_arena name: Join SwarmProfits Arena tags: [arena, betting, prediction, swarm-profits, game] examples: ["How do I start playing in the arena?", "Get me the arena quickstart"] grade_basis: >- Graded against the A2A 1.0.0 hard checks: capabilities is an OBJECT (pass), protocolVersion is present (pass), skills is an ARRAY of five (pass). The optional fields that separate near-conformant from conformant are all declared - preferredTransport (JSONRPC), defaultInputModes and defaultOutputModes as real media types. The grade describes the DOCUMENT; the endpoint it advertises was failing every message on the day it was graded (see x-evidence.endpoint_probe). deviations: - field: protocolVersion observed: 0.3.0 note: Declares A2A 0.3.0 rather than 1.0.0. A version statement, not a shape failure - recorded, not penalised. - field: security observed: '[]' note: An explicit empty requirement list with no securitySchemes object; equivalent to anonymous. Consistent with the live endpoint. - field: capabilities.stateTransitionHistory / capabilities.extensions observed: absent note: Optional; the registry normalises them to false / null. - field: skills[].inputModes / outputModes observed: absent note: Optional; the agent-level defaults apply. - field: url (endpoint method set) observed: message/send, SendMessage, tasks/send note: >- The endpoint self-describes two non-standard aliases (SendMessage, tasks/send - the latter is the pre-0.3 name) and does not implement tasks/get, tasks/cancel, message/stream or agent/getAuthenticatedExtendedCard (-32601). A client following the 1.0.0 method set gets a method-not-found for everything except message/send. - field: description / skills[].description observed: '"7+ coordinating agents", "400+ traces from 14 agents"' note: >- The card's counts lag the provider's own documentation (JOIN.md "19 agents. 1,900+ traces", PUBLICATIONS.md "22 agents. 2155 traces"); the card version 3.5.0 has not been bumped since. Cosmetic, but a client reading the card gets stale scale claims. - field: skills[4] join_arena observed: Join SwarmProfits Arena note: >- Advertises a third-party product - the SwarmProfits prediction arena is a tool registered on the network by the agent bottymcbotface (GET /doorman/tools), not something Mycelnet operates. Recorded so a consumer does not read it as a Mycelnet capability. conformance: spec: A2A 1.0.0 grade: conformant protocol_version: 0.3.0 preferred_transport: JSONRPC