Extentos MCP Server

Give your AI coding agent the tools to build smart-glasses apps.

Extentos is an MCP (Model Context Protocol) server that lets agents like Claude Code, Cursor, and Cline add smart-glasses capabilities — camera capture, voice triggers, live transcription, audio playback — to any Android or iOS app. The tools are deterministic: discovery, scaffolding, validation, and a browser-based simulator for end-to-end testing without hardware.

Works with Meta smart glasses today (Ray-Ban Meta, Oakley Meta, Meta Ray-Ban Display), with a multi-vendor architecture by design.

This repository is the public home of @extentos/mcp-server — releases, changelog, and issue tracking. The package ships on npm.

Install

Claude Code

claude mcp add extentos -- npx -y @extentos/mcp-server@latest

Cursor — one click:

Cline / any MCP client

{
  "mcpServers": {
    "extentos": {
      "command": "npx",
      "args": ["-y", "@extentos/mcp-server@latest"]
    }
  }
}

Requires Node 20+.

Related MCP server: 0nMCP

What your agent can do with it

The workflow an agent typically runs: discover capabilities → scaffold the connection module → write handler code against the SDK → validate → simulate end-to-end → production checklist.

Tools

• getPlatformInfo — SDK capability catalog, version/artifact info, and platform constraints

• getCapabilityGuide — per-feature call shapes in Kotlin + Swift with gotchas

• getCodeExample — canonical end-to-end patterns (voice Q&A, photo-describe, live transcription UI, …)

• getMigrationGuide — mapping table and cutover plan from Meta DAT direct usage to Extentos

• generateConnectionModule — scaffold the connection UI, manifest, and platform config into your app

• getConnectionPageConfig — read the managed connection-page configuration

• setConnectionPageConfig — update the managed connection-page configuration

• regenerateConnectionPageFile — regenerate a managed connection-page file after config changes

• adoptConnectionPageFile — bring an existing connection-page file under managed configuration

• getAssistantConfig — read the managed voice-assistant configuration (model, voice, memory, wake sound)

• setAssistantConfig — update the managed voice-assistant configuration

• listProjectSounds — list custom sounds uploaded for the project

• addProjectSound — upload a custom sound (earcons, wake sounds) to the project

• shutter — trigger a camera shutter in the active simulator session

• getGatewayUsage — AI-gateway usage and metering for the project

• getCredentialStatus — which credentials (Meta DAT, AI provider) are configured

• setCredential — store a credential for the project

• getProjectAnalytics — usage analytics for the project's glasses integration

• getVoiceCommandGuidance — implementation guidance for voice triggers and wake phrases

• getPermissions — the exact permission set the integration needs per platform

• inspectIntegration — read the current state of the integration in your repo

• validateIntegration — deterministic checks of the integration (manifest, config, wiring)

• createSimulatorSession — mint a browser-simulator session for end-to-end testing

• ensureSimulatorBrowser — open/attach the simulator browser tab for the session

• completeAuthLink — finish the device-code sign-in flow

• getEventLog — read the session event log (transport, audio, camera, speak, toggle, stream, system)

• injectTranscript — inject a voice transcript into the running app, as if the wearer spoke

• injectAssistantUtterance — drive the voice assistant with an utterance end-to-end (real provider session)

• assertToolCalled — wait for an assistant tool-call event matching name/args (agent E2E loop)

• getSimulatorStatus — live session state: phase, roles, open streams, freshness, test videos

• setSimVideo — pipe a test video into the simulated camera

• setSimDevice — switch the simulated glasses model

• getDisplayState — read the rendered display tree and interactive element ids (Ray-Ban Display)

• injectInput — drive display input: select/navigate/back, like Neural Band gestures

• injectHardwareButton — simulate the hardware capture button

• getProductionChecklist — the path from simulator to real glasses

• getCredentialGuide — how to obtain and configure Meta DAT credentials

• searchDocs — search the bundled, versioned documentation topics

How it fits together

• Your app depends on the native SDK: com.extentos:glasses (Android, Maven Central) or the Swift package (iOS)

• Your code subscribes to capability primitives from its own handler classes — glasses.audio.transcriptions(), glasses.camera.capturePhoto(), glasses.audio.speak()

• The simulator is the same app on a different substrate: WebSocket transport instead of Bluetooth, so agent-verified behavior carries to hardware

Discovery, validation, and guidance tools work anonymously. Creating browser-simulator sessions and scaffolding require a free account — the server walks the agent (and you) through a device-code sign-in when needed.

Links

• Documentation

• Getting started with an agent

• MCP tools reference

• The smart-glasses ecosystem reference

Issues & feedback

Bug reports and feedback are welcome — open an issue. The package is pre-1.0; tool surfaces may evolve between minor versions.

For security reports, see SECURITY.md — please don't open public issues for vulnerabilities.

License

MIT