apple-event-mcp

MCP server and webhook worker for Apple event liveblog updates.

It pulls WWDC26 coverage from public liveblog pages and exposes structured posts to AI agents. It can also poll continuously, match posts against user interests, and send signed webhooks.

Sources

• Engadget: schema.org LiveBlogPosting, structured posts, per-post images
• MacRumors: static live coverage, timestamped posts, keynote images
• iClarified: dense transcript-style keynote feed
• Macworld: commentary-style liveblog entries

Install

npm install
npm run build

MCP

Run the stdio MCP server:

npm run dev:mcp

Built command:

npm run build
node dist/mcp/server.js

Example MCP client config:

{
  "mcpServers": {
    "apple-event": {
      "command": "node",
      "args": ["/absolute/path/to/apple-event-mcp/dist/mcp/server.js"]
    }
  }
}

Tools:

• list_events
• list_liveblog_sources
• get_liveblog_posts
• search_liveblog_posts
• get_event_summary
• get_noteworthy_updates

Resources:

• apple-event://wwdc2026/live
• apple-event://wwdc2026/summary

Webhook Worker

Copy and edit the config:

cp config.example.json config.json

Run once:

npm run dev:worker -- --config config.json --once

Run continuously:

npm run dev:worker -- --config config.json

Replay Mode

Replay mode simulates a live event without network access. Batch mode reads fixture posts and reveals a few more posts each poll cycle.

Run the first cycle:

APPLE_EVENT_INTERESTS="Siri,Xcode" \
APPLE_EVENT_STORE_PATH=/tmp/apple-event-seen.json \
npm run dev:worker -- --replay --replay-reset --once

Run the next cycle:

APPLE_EVENT_INTERESTS="Siri,Xcode" \
APPLE_EVENT_STORE_PATH=/tmp/apple-event-seen.json \
npm run dev:worker -- --replay --once

Useful replay settings:

• replay.mode: batch or timed
• replay.fixturePath: JSON array of LiveblogPost objects
• replay.batchSize: number of additional fixture posts revealed per poll
• replay.statePath: JSON file tracking the current replay cycle
• replay.timeScale: timed replay multiplier, for example 120 means 1 real second equals 2 event minutes
• --replay-reset: reset the replay cycle counter

Replay still uses the normal seen-post store, so only newly revealed matching posts notify.

Timed replay from captured posts

For a more realistic no-live-event test, capture crawlable WWDC posts locally, then replay their original timestamps against the current clock. Captured publisher content is written under fixtures/local/, which is ignored by git.

npm run capture:replay -- --source all --limit 300 --out fixtures/local/wwdc26-live.replay.json
npm run build

Start the simulation:

APPLE_EVENT_INTERESTS="Siri,Xcode" \
APPLE_EVENT_STORE_PATH=/tmp/apple-event-seen.json \
APPLE_EVENT_REPLAY_STATE_PATH=/tmp/apple-event-replay.json \
APPLE_EVENT_REPLAY_FIXTURE=fixtures/local/wwdc26-live.replay.json \
APPLE_EVENT_REPLAY_TIME_SCALE=120 \
node dist/worker/poller.js --replay --replay-mode timed --replay-reset --once

Run again without --replay-reset to advance the simulated event:

APPLE_EVENT_INTERESTS="Siri,Xcode" \
APPLE_EVENT_STORE_PATH=/tmp/apple-event-seen.json \
APPLE_EVENT_REPLAY_STATE_PATH=/tmp/apple-event-replay.json \
APPLE_EVENT_REPLAY_FIXTURE=fixtures/local/wwdc26-live.replay.json \
APPLE_EVENT_REPLAY_TIME_SCALE=120 \
node dist/worker/poller.js --replay --replay-mode timed --once

Webhook requests are JSON POSTs. If a target has secret, requests include:

x-apple-event-signature: sha256=<hmac>

Environment

• APPLE_EVENT_INTERESTS: comma-separated interests
• APPLE_EVENT_POLL_SECONDS: polling interval override
• APPLE_EVENT_STORE_PATH: JSON dedupe store path
• APPLE_EVENT_REPLAY_FIXTURE: replay fixture override
• APPLE_EVENT_REPLAY_MODE: batch or timed
• APPLE_EVENT_REPLAY_BATCH_SIZE: replay batch size override
• APPLE_EVENT_REPLAY_STATE_PATH: replay state path override
• APPLE_EVENT_REPLAY_TIME_SCALE: timed replay speed multiplier
• APPLE_EVENT_REPLAY_START_AT: fixed simulated start timestamp
• APPLE_EVENT_REPLAY_ORIGINAL_START_AT: fixed source-event start timestamp

Notes

These sources are public webpages, not stable APIs. Adapters are intentionally small and covered by parser tests so they can be updated quickly when page markup changes.