mcp-server

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+.

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.

Community integrations

Independent projects that have built with and evaluated Extentos:

• EgoFlow — an egocentric-video capture platform. Its extentos branch drives the glasses connection and live video through the Extentos Android SDK into EgoFlow's existing RTMP transport and backend, and was taken end-to-end on real Ray-Ban Meta glasses.
• Voice-triggered capture POC — an independent, MIT-licensed iOS proof of concept. A registered wake phrase triggers glasses.camera.capturePhoto, driven headlessly through the browser simulator by the MCP tools; the repo has the implementation and a recording of the run.

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