Contribute & MCP

Plague Tracker exposes a public Model Context Protocol server. AI agents and assistants can use it to read the scored data, including the spread map, and to send in hints, evidence and symptom reports.

For AI agents: read /skills.md. It covers how to query, contribute and subscribe, in Agent Skill format (it can be installed as a skill).

Endpoint

https://plaguetracker.com/mcp

Streamable HTTP transport, stateless. No key is needed to read or submit. Submissions without a key go to the review queue.

Claude Code

claude mcp add --transport http plaguetracker https://plaguetracker.com/mcp

Other MCP clients (JSON config)

{
  "mcpServers": {
    "plaguetracker": {
      "type": "http",
      "url": "https://plaguetracker.com/mcp"
    }
  }
}

Partner keys

Research groups and newsrooms that submit regularly can request an API key, sent as Authorization: Bearer <key>. Trusted keys skip the review queue, but their submissions are still scored like everyone else's.

Tools

Read

• get_situation — Key figures with ranges and confidence, counts, top claims, health guidance
• list_claims — Filter claims by type, verification, likelihood, location, date or text
• get_claim — One claim with all of its evidence and sources
• get_spread_map — GeoJSON of affected locations; pass `until` to replay the spread
• get_timeline — Day-by-day claims and locations
• get_symptom_overview — Aggregated, reviewed symptom reports per locality
• list_locations — Known places (use their ids when submitting)
• list_sources — Every cited source with type and reliability

Submit

• submit_hint — New information: a case, death, measure, statement or rumor, with source links
• submit_evidence — Add a supporting or contradicting source to an existing claim
• report_symptoms — People with symptoms: what, when, where (town level), exposure. No personal data.

Every submission must identify the submitter: an AI agent (name and model) or a human. Provenance is stored with the submission and shown on the public site ("submitted by an AI agent").

Follow changes (events)

Every change to the public data creates an event: claim.published, claim.score_changed, evidence.added,key_figure.changed and symptom_reports.updated (aggregates only). Only published data is ever sent out. There are three ways to receive them:

• MCP subscriptions. In a session, subscribe to plaguetracker://events, plaguetracker://situation orplaguetracker://claims/<id>. The server pushes notifications/resources/updated, then you read the resource.
• Polling. Call get_events with the last event id you saw (after_id). This works with every MCP client.
• Webhooks (API key required). Call create_webhook with an https URL, optionally filtered by event type or claim ids. We POST {id, type, at, claim_id, data} and sign it in X-Plaguetracker-Signature: t=<unix>,v1=<hex>, where v1 is HMAC-SHA256(secret, t + "." + body). Failed deliveries are retried with backoff.

Without MCP, use the plain SSE stream (it resumes with Last-Event-ID):

curl -N "https://plaguetracker.com/api/events/stream?types=evidence.added,claim.score_changed"

Rules for submissions

• Link sources. A hint without a link starts at very low likelihood, and all unsourced tips together count as a single source.
• Search first. If the claim already exists, add evidence to it instead of creating a duplicate.
• Quote, don't paraphrase, when you attach evidence. The original language is fine.
• No personal data. No names, phone numbers, addresses or photos of private individuals. Give locations at town or district level only.
• Don't fabricate. Submissions are logged and reviewed, and repeated low-quality or false submissions get a client blocked.

Symptom reports

Reports go into a review queue and are only ever published as anonymous counts per locality. This is not a medical service. If someone is ill after possible exposure, call 112 or 103 first.

Plain JSON API

• /api/situation.json
• /api/claims.json (?type=&verification=&min_likelihood=&location=&since=&until=&q=)
• /api/claims/<id>.json
• /api/map.geojson (?until=YYYY-MM-DD)
• /api/timeline.json
• /api/changes.json (score changes over time)
• /api/events/stream (SSE, ?types=&claim_id=&after_id=)