MCP server & AI agents

Geolocated world events from global news media, AI-enriched with severity/GEOINT scores and collection recommendations.

Geolocated world events (geopolitical, security, disaster, infrastructure) distilled from global news media and AI-enriched with severity/GEOINT scores and collection recommendations. Filter by bbox, date window and category.

Result (structuredContent)

AI world brief: daily (prior UTC day, all plans) or weekly/monthly by plan. freshness.is_stale: no newer day, so relay as stale.

The World Brief — an AI digest of worldwide event signals (headline, executive summary, top developments, per-theme roll-up, ranked signals). period=daily (default) covers the previous UTC day and is open on every plan; weekly and monthly are opened by the key owner's plan. If freshness.is_stale, a newer day is not yet available, so relay it as possibly out of date.

Result (structuredContent)

The calling key's remaining token balance and plan capabilities. Pre-flight a metered call with it.

The calling key's remaining token balance and plan capabilities. Pre-flight a metered call with it.

No arguments.

Result (structuredContent)

Search the imagery catalog over an area and window. Metadata only, no pixels. eventDate tags each scene pre/post — **a same-day scene is same_day_unknown, never post** — and eventPoint/eventAoi add target_relation, so a scene that only clips the bbox is not read as covering the event.

Search the imagery catalog (Sentinel-1, Sentinel-2, NISAR L-band) over an area and window. Metadata only, no pixels. With eventDate each scene is tagged timing=pre/post/same_day_unknown — **a same-day scene is same_day_unknown, never post**, without a real event time — and meta reports bracketing and sar_pair_status. eventPoint/eventAoi add target_relation, so a scene that only clips the wide bbox is not read as covering the event.

Result (structuredContent)

The deterministic imagery plan for ONE event: checks BOTH sensors exactly once — sentinel-1-grd (SAR, the only look that survives cloud and night) and sentinel-2-l2a — against the event footprint and a pre/post window. **A VHR recommendation never invalidates what the free catalog showed.**

The deterministic imagery plan for ONE event: the server checks BOTH sensors exactly once — sentinel-1-grd (SAR, the only look that survives cloud and night) and sentinel-2-l2a — against the event footprint and a pre/post window, reporting how many scenes COVER the event and are usable, SAR pair status and bracketing. **A VHR recommendation never invalidates what the free catalog already showed.** An event with no resolvable footprint is refused, not guessed.

Result (structuredContent)

WHERE, and with what class of satellite, observation is most worthwhile now: composite IMPORTANCE crossed with the SPEC CLASS the required resolution demands — coarse, hr (free Sentinel-class) or vhr. Deterministic.

WHERE, and with what class of satellite, observation is most worthwhile now. Crosses each event's composite IMPORTANCE with the SPEC CLASS its resolution demands: coarse (<=100 m), hr (<=10 m, free Sentinel-class) or vhr (sub-metre, commercial). Deterministic, no LLM.

Result (structuredContent)

WHEN a place can next be imaged and by WHAT: 13 free-systematic and commercial-taskable families. **Every pass is a GEOMETRIC access opportunity** — no operator plan is consulted. retrieval_ok:false means timing is UNAVAILABLE, never "no passes".

WHEN a place can next be imaged and by WHAT. SGP4 over day-cached elements for 13 families: free SYSTEMATIC (Sentinel-1/2, Landsat, NISAR) and commercial AGILE (WorldView, ICEYE, Capella, SkySat, Umbra, Synspective, iQPS, RADARSAT-2, COSMO-SkyMed — taskable access needing a paid order, NOT a guaranteed collect). **Every pass is a GEOMETRIC access opportunity** — no operator plan is consulted. retrieval_ok:false means timing is UNAVAILABLE, never "no passes".

Result (structuredContent)

AI remote-sensing deep-dive for one signal: what to observe, sensors, a collection window. Also returns a deterministic `context` whose `imagery_handoff.parameters` are the exact search_imagery inputs. Cached per signal; not-observable signals are rejected before any charge.

AI remote-sensing deep-dive for one signal: what to observe, recommended sensors, a collection window. Also returns a deterministic `context` whose `imagery_handoff.parameters` are the exact search_imagery inputs for real pre/post candidates. Cached per signal, not re-charged. Signals that are not satellite-observable are rejected BEFORE any charge.

Result (structuredContent)

Ask the Delta Analyst an OSINT/GEOINT question; returns a structured brief. **Durable async**: {status:"processing", job_id} comes back immediately and the run finishes in a background worker. Fetch with get_analyst_job, or re-send the SAME idempotencyKey (no second charge).

Ask the Delta Analyst an OSINT/GEOINT question. Agentic multi-step analysis returning a structured brief (summary, findings with collection recommendations, assessment, citations). **Durable async**: returns {status:"processing", job_id} immediately and finishes in a background worker, so it survives a client timeout. Fetch with get_analyst_job, or re-send the SAME idempotencyKey (no second charge).

Result (structuredContent)

Status and result of an ask_analyst run. Status is "processing", "done" or "error"; result_quality.status is **SEPARATE** — a done job can carry a partial result.

Status and result of an ask_analyst run by job_id. Status is ONLY "processing", "done" or "error". result_quality.status ("complete" | "partial") is **SEPARATE from job status**: a done job can carry a partial result. Visible only to the key that created it.

Result (structuredContent)

What actually CHANGED about the events in an area, not which articles are new. Each change is labelled `world` (the event's own state moved) or `measurement` (what we can see moved). Plan-dependent values say so (`locked_by`).

What actually CHANGED about the events in an area, not which articles are new. Each change carries what it was and what it became, labelled `world` (the event's own state moved) or `measurement` (what WE can see moved: location resolved, imagery arrived, SAR pair ready). The right source for "what is new since last time".

Result (structuredContent)

One event end to end: state plus every change in order — the "new event or update" distinction a feed cannot make. `occurred_at_basis` distinguishes stated, absent and never measured. Plan-dependent: see `locked_by`, `plan_lock`.

One event end to end: current state and every change in order — the "new event or update to an old one" distinction a feed cannot make. Returns the canonical event (where, when it HAPPENED as distinct from when it was reported, casualties, attribution and its grounding), a timeline and every source. `occurred_at_basis` distinguishes stated, absent and never measured.

Result (structuredContent)

The Watchlist as one list, each entry with a state bucket. `updated_since` returns only watches whose CONTENT changed after that instant — what a synchronised copy should poll.

The Watchlist: measured areas, event-watched areas and tracked events as one list, each with a state bucket (needs_attention, changed_today, awaiting_data, stable). A watch groups everything on the same target. `updated_since` returns only watches whose CONTENT changed after that instant — what a synchronised copy should poll.

Result (structuredContent)

What changed across the Watchlist, newest first: one row per change a watch reported, with read state. Reading does not mark them read. A value the plan does not open is `locked`, never filled in.

The Notification Center: every change a watch reported — an event development, a newly related event, an anomaly in a measurement, a new event at a site, a new scene, a fired standing order — one row each, newest first, the same rows as the bell in the app and the Watch update emails. `unread_count` in meta is the total unread regardless of filters. **Listing does not mark anything read**: the unread state belongs to the user's inbox, and a conversation reading it should not clear the bell. Marking read is POST /api/v1/notifications/read or client.notifications.mark_read in the Python SDK — deliberately not an MCP tool (inbox housekeeping, not something a conversation needs). A value the plan does not open is withheld: `locked` is true and `lock_plan` names the plan that opens it. For one watch in full, follow watch.id into get_watch.

Result (structuredContent)

One watch end to end: target, state, latest change, measurements with recent series, standing-order questions, and for an event watch its stage, developments, imagery check and thread, plus this account's notes.

One watch end to end: target, state, latest meaningful change, measurements with recent series, standing-order questions, and for an event watch its stage (how precisely it is located), developments, imagery availability and full thread — plus this account's notes and judgments. An event watch reports signal.imagery_check: whether imagery is being checked for it and, if not, why (see create_watch for the rule and the charge). The FULL measurement history is on the REST API (GET /api/v1/monitoring/{areaId}) and the Python SDK (client.monitoring.get).

Result (structuredContent)

Add a target to the Watchlist. EVENT: the signal's event id binds the canonical event — **never watch an article URL**. AREA: a bbox. SITE: entity_id or lat+lon, and scale. Free to create; where the plan includes the collection section, a watched event's imagery is then checked at 1 token a day (up to 14 days).

Add a target to the Watchlist. EVENT: pass the signal's event id and the server binds the canonical event, so later reporting and even a cluster merge stay on the same watch — **never watch an article URL**. AREA: pass a bbox of at most 100,000 km² (larger is refused with 400 invalid_input; for a whole country use SITE at scale "country"); measuring there and a standing order on its events are created over the REST API or the Python SDK (monitoring / standing-orders) and auto-join the same watch. SITE: watch a specific place for change — a registry facility (entity_id from search_entities) or any point (lat, lon) — at a scale: "facility" (radius 250 m–3 km; measured from imagery by default when the kind of facility has a suggested measurement, e.g. vessel counts for a port), "city" (5–20 km; measured only when you pass a metric) or "country" (reporting only; never measured). Linked events come back with how they were linked (named / nearby / country). A refused explicit metric creates nothing; a refused default is reported in measurement.reason. Measuring uses a monitored-area slot and costs the same as a monitored area. IMAGERY FOR A WATCHED EVENT: creating the watch is free. Where the account's plan includes the collection section (the imaging recommendation), imagery of a watched event is then checked automatically at 1 token a day per watch, until before and after scenes are found or 14 days after the first report; a day without enough tokens is skipped, not charged. On other plans imagery is not checked and nothing is charged. Imagery is checked only for watched events — an event that is not watched carries no imagery status.

Result (structuredContent)

Pause, resume or close a watch. **Pausing is not a display state**: monitored areas stop being measured and charged, standing orders stop checking. **Closing states how the question ended** (close_reason).

Pause and resume a watch, CLOSE it when the question is settled, or turn on `imagery_alerts`. **Closing states how the question ended** and requires `close_reason`: `resolved` (an answer was reached), `lapsed` (interest moved on without an answer) or `false_alarm` (never a watchable event). A closed watch stops notifying, stops its bound measurements and standing orders, and frees an active slot — so close rather than delete when the record is worth keeping. **Pausing is not a display state over live billing**: monitored areas stop being measured and charged, standing orders stop checking. Resuming restarts them, standing orders due immediately. Use it instead of delete_watch when the history is worth keeping. `imagery_alerts` is for AREA watches and facility- or city-scale SITE watches: it reports that a NEW scene now covers the area — availability, never what is in the scene. For an event, post-event coverage already arrives as a development on the event itself. `notify_chat` posts the watch’s notifications to the Slack or Discord channel the user connected in account settings (connecting happens in the app); with none connected the setting is kept and nothing is posted.

Result (structuredContent)

Delete a watch and its underlying resources — bound monitored areas with their history, and standing orders. To stop without losing them, pause with update_watch.

Delete a watch and its underlying resources — bound monitored areas with their measurement history, and standing orders — exactly as the in-app Watchlist does. To stop without losing anything, pause with update_watch instead.

Result (structuredContent)

Add a note to a watch, start a thread, or reply. **Confidence is what separates a note from a judgment**: with one it is a JUDGMENT that supersedes the previous judgment rather than overwriting it.

Add a note to a watch, start a thread, or reply — one append-only ledger. A title makes it a thread; replies are one level deep. **Confidence is what separates a note from a judgment**: with one it is a JUDGMENT that supersedes the previous judgment rather than overwriting it. It does not change the event's public record.

Result (structuredContent)

Find a named place (airport, base, plant, port, dam, strait…) in any language or by IATA/ICAO code. Max 50.

Find a place in the location registry — airports, military and naval bases, launch sites, nuclear facilities, ports, power plants, dams, bridges, government sites, chokepoints and named seas. Matches names in many languages and IATA/ICAO codes, most prominent first. At most 50 rows, no pagination. Cite the returned attribution.

Result (structuredContent)

What has happened at one place: the registry record plus its most recent linked events, each with HOW it was linked. **`link_kind: proximity` is an association, not a statement that the event happened there.**

What has happened at one place: the registry record plus its most recent linked events (`event_count` is the total), each carrying HOW it was linked — `link_kind` `named_location` (a report named this place as where it happened), `named_target` (named it as the target), `measured` (a satellite measurement recorded near it, with `distance_m` — what was measured, not a report of what happened) or `proximity` (located within range, with `distance_m`). **Read the basis: proximity is an association, not a statement that the event happened there.**

Result (structuredContent)

The relations recorded for one event: others at the same registry facility, naming the same place, or stored as the same campaign, each saying what is shared; plus reports merged into it. Merely-nearby events are not relations and are not returned. Plan-dependent: see `plan_lock`.

The relations recorded for one event — nothing is inferred on request. **Every entry in `related` says what the two events share** in `shares`: the same registry facility, the same named place (resolved to a town or finer), or a relation stored when the events were processed (`same_campaign`). Reports of the same event merged into it are in `merged_in`. Events that are merely close in space and time are not relations and are not returned. Each facility and place lists its most recent other events; `other_count` is the total.

Result (structuredContent)