Daily digest of GitHub star activity across every non-archived wyre-ai
repo, posted to Slack.
Runs every day at 14:00 UTC (10am ET / 9am ET depending on DST) via GitHub Actions.
Why no download counts? npm downloads aren't queryable for our packages — we publish to GitHub Packages (
npm.pkg.github.com), which has no public counter, and the public npm registry reports 0. GitHub deprecateddownload_counton container packages in late 2022. The MCP Registry and Glama.ai are catalogs, not analytics services — they expose presence and version, not downloads. So this digest tracks reach and freshness across distribution channels, not downloads or traffic. Adoption metrics (tool calls, active orgs, top vendors) live in a separate digest sourced from the gateway DB — seewyre-ai/adoption-watcher.
-
Delivery uses the shared WYRE Notifier Slack app (WYRE AI workspace, manifest in
wyre-technology/.github→slack-app/notifier/) via the org-levelSLACK_NOTIFIER_BOT_TOKENActions secret — nothing to set per repo. The target channel isSLACK_CHANNEL_IDin.github/workflows/daily.yml(currentlyC0BR5S8F6BZ, #github-activity). -
(Optional) If the default
GITHUB_TOKENlacks read access to private repos you want included, add a fine-grained PAT asGH_API_TOKENwith:Repository → Administration: Read,Metadata: Read. Without it, the digest will only see public repos under the org. -
Glama.ai directory coverage needs an API key. Create one at https://glama.ai/settings/api-keys and store it as an Actions secret named
GLAMA_API_KEY(repo or org — the daily workflow readssecrets.GLAMA_API_KEY). Requests send it asAuthorization: Bearer. If the secret is missing or Glama rejects it (HTTP 401/403), the digest says Glama was skipped instead of listing every server under "Not on Glama". -
Trigger manually the first time to confirm formatting:
gh workflow run daily.yml --repo wyre-ai/stars-watcher
script/report.py is plain stdlib Python — no pip install. It hits:
GET /orgs/wyre-ai/repos(paginated) — starsGET /repos/wyre-ai/<repo>/traffic/clones— 14-day clones (needs theGH_API_TOKENPAT; skipped gracefully without it)GET /repos/wyre-ai/<repo>/releases/latest— release tagsGET registry.modelcontextprotocol.io/v0/servers— MCP Registry coverageGET glama.ai/api/mcp/v1/servers— Glama.ai directory coverage (requiresGLAMA_API_KEY; skipped with an explicit Slack line when the key is unset or rejected). Results include a Glama.ai attribution link, which their API data license requires when the directory data is shown.
The registry and Glama sections track reach and freshness — whether each
*-mcp server is published, current, and indexed. See
docs/superpowers/specs/2026-05-20-registry-reach-tracking-design.md.
PulseMCP traffic tracking (estimated visitors over the last four weeks) was
removed — the free public search endpoint it depended on
(www.pulsemcp.com/api/v0.1/servers) is gone. PulseMCP now runs a
private, API-key-gated partner API at api.pulsemcp.com; re-add this section
if WYRE gets partner credentials from hello@pulsemcp.com.
The result is written to state/snapshot.json and diffed against the previous
day's snapshot. Deltas, top movers, and totals are formatted as Slack Block Kit
and POSTed to the webhook. The new snapshot is then committed back so tomorrow
has something to diff against.
Stdlib unittest, no pip:
python -m unittest discover -s tests -v
Free. Runs in the GitHub-hosted Linux runner under the standard org allotment; the script is short and finishes in under a minute. The repo is itself only producing one commit per day.
Tweak script/report.py. Want it to also pull PyPI stats? Add a pypistats
fetcher next to the npm one. Want a weekly summary alongside the daily? Add a
second cron + a separate snapshot file.
The Slack message is Block Kit JSON — visualise edits with the Block Kit Builder.