What we tested
Probed live on 26 Sep 2026 against rubric v1.2. 1 of 5 checks passed. 2 unknown.
PROBE-VERIFIEDtier 2 of 3
—MCP server card/.well-known/mcp/server-card.json · absent
not scored —Platform-suppliedno
not scored ✓Other agent files at well-known pathsagent-skills/index.json · oauth-authorization-server · oauth-protected-resource · http-message-signatures-directory · found: OAuth authorization server metadata, OAuth protected resource metadata
not scored What its MCP server offers
We connected, ran tools/list, and recorded what came back on 26 Sep 2026. 19 tools.
list_voicesspeaksend_voiceinboxtranscribekoe_narratekoe_check_readingkoe_bangumikoe_my_radiokoe_bangumi_statuskoe_clone_startkoe_clone_submitkoe_clone_statuskoe_speakset_elevenlabs_keyclear_elevenlabs_keykoe_voiceidkoe_voiceid_statuskoe_radio
What an agent could do here
Read from the page’s own content, with the wording each was drawn from. Inference is kept separate from tested facts — an agent completing the action is tier 3.
UNVERIFIED
register“”https://mcp.koe.live/api/auth/request-link
Context
65 sites in the index share the category conversational and voice AI. 2 transactional sites share it.
How we grade
Three tiers. Every listing shows which one it has reached, and when. The rubric.
tier 1Declared
The site says agents are welcome in its robots.txt. A permission, not a capability.
tier 2Probe-verified
We fetched the site and tested each signal ourselves. Dated, repeatable, re-checked on a schedule.
tier 3Behaviourally verified
An agent completed the action end to end — booked, bought, submitted — and it worked.
Own this site?
2 checks did not pass when we last looked: llms.txt, API catalog. Each is a small, specific fix; the rubric says exactly what we look for.
Behavioural verification — an agent completing your booking, purchase or contact flow end to end, with the date published here — opens to a small founding cohort first. Leave an email to be in it.
Related verified sites
Wispr FlowAPI documentation for Wispr Flow, a voice interface platform providing voice-to-text transcription, real-time audio streaming, and voice dictation features via REST and WebSocket APIs.
conversational and voice AI Kallglot APIAPI for building real-time multilingual voice applications with translation sessions, AI voice agents, live transcription, conversation analysis, and telephony integration (Twilio, Telnyx, SIP).
conversational and voice AI Mihu API DocumentationAPI documentation for Mihu platform enabling AI-powered voice and text conversation systems, contact and campaign management, webhook integration, and automated multi-channel messaging (voice, WhatsApp, SMS, email).
conversational and voice AI 60db60db provides a modern voice AI platform with APIs and SDKs for text-to-speech, speech-to-text, AI chat, memory management, custom voices, music generation, and phone call management.
conversational and voice AI AutocallsAutocalls.ai provides a platform to build AI voice assistants for automating inbound and outbound phone calls, call campaigns, appointment scheduling, lead management, customer support, and integrations via API or webhooks.
conversational and voice AI Avaturn.LiveDocumentation for Avaturn.Live, providing real-time conversational video avatars for web applications, including voice-to-voice avatar interactions, session lifecycle management, and integration with conversation engines like OpenAI Realtime and Cartesia.
conversational and voice AI BosonAPIs for live voice conversation, text-to-speech synthesis, and talking-avatar video generation, including Higgs Realtime, Higgs TTS 3, and Higgs Avatar, with documentation and usage-based pricing.
conversational and voice AI DeepslateDocumentation site for Deepslate's voice AI platform, including an end-to-end speech-to-speech AI model, real-time voice conversation tools, SDKs, API references, and integration guides for building voice AI applications.
conversational and voice AI Record generated from a live database query. Method: about the index.
How we found this site: probe discovery. This site didn’t declare agent access — we found it by checking the open web. Everything above was tested the same way, against the same rubric, on the date shown. How we choose what to check, and how to opt out →