Eavesy

Drop anything in, get clean Markdown context out for Claude, Cursor, ChatGPT, or any AI agent.

Install

Mac app (Apple silicon, macOS 13 Ventura or later):

brew install --cask williamrogersdev/eavesy/eavesy

Or download the DMG, open it, and drag Eavesy into Applications. The app updates itself.

Web app: eavesy.com works in any browser, no install.

Connect your agent: create an MCP connection in the web app's Settings, then paste the config it shows into your agent:

{
  "mcpServers": {
    "eavesy": {
      "type": "http",
      "url": "https://mcp.eavesy.com",
      "headers": { "Authorization": "Bearer YOUR_TOKEN" }
    }
  }
}

See Installing the Mac app and Connect an agent over MCP for permissions and the available tools.

What it is

Eavesy turns what you drop into it into clean, structured Markdown context that is quick to copy into Claude, Cursor, ChatGPT, or any agent, or to reach over MCP. It is local-first: your captures live on your disk as Markdown and JSON.

What ships today:

• Mac app (desktop/): from the launcher, press ⌘V to save text or an image from your clipboard, write a note or take a screenshot (once you sign in, these sync to your other Macs), or dictate and record (these upload for transcription once you sign in) your screen, system audio, and microphone. Everything is mirrored to ~/Eavesy/Recordings as Markdown, JSON, and media files.
• Web app (web/): dashboard, browser recorder, search, summaries and transcripts you can copy or export as Markdown, and settings, including connecting your own OpenRouter account.
• Hosted MCP server (workers/mcp): tools to search and read everything in your account (recordings, dictations, notes, screenshots and other captures) and fetch summaries, transcripts and action items. With write access, agents can also save notes, tick off action items, and rename or file items under projects. The same tools are a REST API at mcp.eavesy.com/v1 (OpenAPI at /v1/openapi.json), and signed webhooks fire when a recording finishes or a dictation, note or capture is saved.

Meeting recording is one capability, not the whole product. Not shipped yet: forwarding messages and orders from other apps, live captions, and the cross-device "drops" work. See the roadmap.

Eavesy is free to use. Everyone gets the same daily limits; see Fair-use limits for the numbers and how to change them when self-hosting.

Develop

Production architecture

Concern	Implementation
Web app	TanStack Start + React 19 on Cloudflare Workers
Authentication	Clerk
Relational data	Cloudflare D1 + Drizzle
Audio	Cloudflare R2
Segment processing	Cloudflare Queues
Transcription	Workers AI Whisper Large v3 Turbo
Summaries	Anthropic
Semantic search	Workers AI embeddings + Vectorize
Hosted MCP	Streamable HTTP Worker at mcp.eavesy.com
macOS app	Tauri 2 + Rust + Swift/ScreenCaptureKit

Long recordings are divided into roughly four-minute AAC/M4A segments, with a five-minute hard cap. Each segment is persisted locally before it is uploaded. The processing worker transcribes segments asynchronously, periodically updates rolling notes, and generates the final summary and embedding after recording stops. D1 stores session state; no live-recording Durable Object is required.

Repository

eavesy/
├── web/                 # TanStack Start web app and authenticated APIs
├── workers/
│   ├── processing/      # Queue consumer, finalization, and cleanup cron
│   └── mcp/             # Hosted MCP server and REST API (/v1)
├── desktop/
│   ├── launcher-ui/     # Offline launcher UI bundled with Tauri
│   ├── src-tauri/       # Rust state, durable uploads, and filesystem mirror
│   └── swift-audio/     # macOS microphone + system-audio capture sidecar
└── docs/
    ├── ARCHITECTURE.md  # How the services fit together
    ├── SELF_HOSTING.md  # Run your own instance; local dev mode
    ├── DEPLOY.md        # Cloudflare deploy runbook
    └── OPERATIONS.md    # Alerts, metrics and recovery

Local development

Prerequisites: Bun, Rust stable and Xcode Command Line Tools. No Cloudflare or Clerk account is needed to run the web app locally:

cd web && bun install
bunx wrangler d1 migrations apply eavesy --local
EAVESY_LOCAL_TEST=1 bun run dev

See docs/SELF_HOSTING.md for what the local mode stubs and how to deploy your own instance on Cloudflare.

Useful checks from the repository root:

bun run build
bun run typecheck
bun run lint
bun run test

cd workers/processing && bun run typecheck && bunx wrangler deploy --dry-run
cd ../mcp && bun run typecheck && bunx wrangler deploy --dry-run
cd ../../desktop/src-tauri && cargo check

Type checks use pinned tsrs in web, both Workers, and the Worker test suite. bun run typecheck:tsc provides the TypeScript fallback; bun run typecheck:bench compares both compilers across all four configurations. See benchmark results.

For the macOS launcher and capture sidecar:

cd desktop
./scripts/prebuild.sh
cd src-tauri
bunx @tauri-apps/cli build --debug --no-bundle

Deploy

Apply D1 migrations before deploying code, then deploy the processing worker, MCP worker, and web worker in that order:

bun run db:apply:remote
bun run deploy:processing
bun run deploy:mcp
bun run deploy

See docs/DEPLOY.md for resource provisioning, secrets, metadata indexes, smoke tests, rollback, and desktop publishing. Desktop build and signing steps are in desktop/BUILD.md.

Release the Mac app

desktop/scripts/build-macos-release.sh
desktop/scripts/release-macos.sh --publish   # notarize, upload to eavesy.com
scripts/publish-github-release.sh            # tag desktop-vX.Y.Z + GitHub Release
packaging/homebrew/bump-cask.sh              # then copy the cask to the tap repo

The Homebrew tap's source lives in packaging/homebrew/.

Acknowledgements

• The macOS launcher panel uses tauri-nspanel (Apache-2.0).

License

Eavesy is licensed under the Apache License 2.0. Versions released before the switch to Apache-2.0 remain available under their original FSL-1.1-ALv2 terms.

Contributions are accepted under the same license; see CONTRIBUTING.md.