Warmth Engine Observatory

Measured conclusion: Deep-tested. L2 outcome: tools_ok.

Coordination Intelligence: AI infrastructure coordination dynamics across geopolitical blocs

Add to your agent

Copy this into your MCP client config:

{
  "mcpServers": {
    "Warmth Engine Observatory": {
      "url": "https://warmthengine.com/mcp",
      "transport": "streamable-http"
    }
  }
}

Measured tools (18)

• search_events read-only

  Search and filter the WEO event corpus by `query` text or classification (bloc, tier, domain, event type, primary industry tag); returns the matching events with their metadata, capped by `limit` (default 20, max 50). Use to discover or shortlist events; for a single event's complete record, use `get_event`. Events enter after a verification lag from their implementation date (six months; 60 days for verified product releases), so recent events may not yet appear, and an empty result is not evidence that no such event has occurred. Free tier returns the classification facts with the deeper assessment and evidence withheld on non-sample events; full tier — and sample events on either tier — return every field.

  Parameters:

  • bloc — string — Filter by bloc: Western, Eastern (tier 3); Western-Led, Eastern-Led (tier 2); Cross-Bloc (tier 1). Each bloc value occurs at one tier only
  • tier — number — Filter by tier: 1 (Cross-Bloc), 2 (Between-Bloc), 3 (Intra-Bloc); each tier carries its own bloc values (see `bloc`)
  • type — string — Filter by event type: Policy, Infrastructure, Corporate, Coordination
  • limit — integer — Number of results (default 20, max 50)
  • query — string — Free-text search across event name and description
  • domain — string — Filter by domain: Technology, Energy/Infrastructure, Security, Trade
  • industry — string — Filter by primary industry tag

  Hints: read-only yes · destructive no · idempotent yes · open-world no

• get_event read-only

  Retrieve one event by ID — classification, assessment, sources, industry tags, Environmental Nexus Tags (ENTs), bloc analysis, and tier rationale. Use when you already hold an event ID; to find IDs, use `search_events`. Pass `event_id` as the zero-padded string (e.g. "06", "141"). `date` is the implementation date (general availability for a product release). Free tier returns the complete record for sample events and the classification facts for all others (deeper assessment and evidence withheld); full tier returns every field.

  Parameters:

  • event_id (required) — string — Zero-padded event ID (e.g., "06", "141")

  Hints: read-only yes · destructive no · idempotent yes · open-world no

• get_connections read-only

  Retrieve Coordination Connections (CCs) — the typed, directional relationships WEO records between events — optionally filtered by event (as source or target), type (1–7), or confidence (CC-V / CC-E / CC-A). For a single CC by its identifier, use `get_connection_by_id`. Each CC carries its id, source/target event IDs, type name + number, confidence, and direction; full tier — and either tier when an endpoint is a sample event — adds the evidence summary, evidence detail, and verification history.

  Parameters:

  • event_id — string — Return only CCs involving this event (as source or target)
  • confidence — string — Filter by confidence level: CC-V, CC-E, CC-A
  • connection_type — number — Filter by CC type (1–7)

  Hints: read-only yes · destructive no · idempotent yes · open-world no

• get_actors read-only

  Retrieve whole Sovereign Capability Profile (SCP) actors — each with its designation (PAA / AIK / ACS / Participant), capability score, severance result, `assessed` (the register month, YYYY-MM, the profile was assessed in), a met/not-met assessment across the seven capability dimensions (D1–D7), and `energy_context`. Use to fetch complete actor records; to pivot one dimension across all actors, compare actors side by side, or read the dimension-watch register, use `query_scp`. Filter by designation or actor name. Free tier returns each dimension's met status and recorded headline (and the US sample in full); full tier adds the per-dimension constraint, qualification, and sources. `energy_context` carries installed generation capacity from the U.S. Energy Information Administration and annual electricity generation from the energy think tank Ember, served in full on both tiers: external reference data outside the WEO analytical stack, national totals, dated observations reported as published, accompanying the profiles but forming no part of them. Descriptive — recorded status, never inference.

  Parameters:

  • actor_name — string — Filter by actor code (ISO 3166-1 alpha-2, exact, e.g. "CN") or name (partial match, e.g. "China"). Named `actor` in `get_capability_links`.
  • designation — string — Filter by designation: PAA, AIK, ACS, Participant

  Hints: read-only yes · destructive no · idempotent yes · open-world no

• get_coverage_stats read-only

  Retrieve the current corpus statistics — live counts and distributions of events, Coordination Connections (CCs), and Sovereign Capability Profile actors across bloc, tier, domain, event type, primary industry, CC type, confidence, and designation, plus Environmental Nexus Tag (ENT) statistics with the Warmth Engine Ratio, counts of the derived layers — Infrastructure Threads, Capability Links with their type split, and Capability Signatures — the event-ID range, and the methodology version. Capability Link and Capability Signature counts are of active records only: retired records are excluded so the figures describe present coverage rather than everything ever assessed, with the active/retired splits published in the two capability tools' metadata. No parameters; figures are computed live, so the response always reflects the present database. The secondary-industry breakdown is full-tier only.

  Hints: read-only yes · destructive no · idempotent yes · open-world no

• get_methodology read-only

  Look up the WEO methodology definition for any platform-specific term, field, or concept (e.g. "CC-V", "T2", "PAA", "ENT-1", "PIET"). Returns the term's definition, its section anchor, a deep link to that section of the published methodology, and the methodology version. Use to resolve any vocabulary the other tools return. Pass `term`; matching is exact-first, then substring, and an unknown term returns a sample of available terms.

  Parameters:

  • term (required) — string — The term to look up (e.g., "CC-V", "T2", "PAA", "ENT-1", "PIET", "Basel Assessment")

  Hints: read-only yes · destructive no · idempotent yes · open-world no

• get_connection_by_id read-only

  Retrieve a single Coordination Connection (CC) by its `connection_id` (e.g. "CC-06-11-3"). Use when you have the exact identifier; to list or filter CCs, use `get_connections`. Returns the CC with full evidence on the full tier — or when an endpoint is a sample event — and the base fields only otherwise.

  Parameters:

  • connection_id (required) — string — CC identifier (e.g., "CC-06-11-3")

  Hints: read-only yes · destructive no · idempotent yes · open-world no

• get_topology read-only

  Retrieve the precomputed Coordination Topology — the structural lineage layer over the parent→child Coordination Connection graph (a DAG). Omit `node_id` for whole-graph network statistics (taxonomy-class counts, roots, multi-parent nodes, dominant roots by coordination reach, weakly-connected components). Pass `node_id` (e.g. "E34") for one node's structural metrics: generational depth (min/max/all-paths), coordination reach, directed betweenness, path diversity, fan-in/out, taxonomy class, component id. For the actual paths between nodes or up/down a lineage use `traverse_coordination`; for one event's value-chain stack use `get_event_stack`. Optional `class` / `cc_type` / `min_reach` filters return matching nodes. The full tier adds the held analyst layer (chain participation, curated orphan/sibling and named-feature sets).

  Parameters:

  • class — string — Filter nodes by taxonomy class: root | internal | leaf | sibling_only | orphan
  • cc_type — number — Filter to nodes incident to a backbone (parent→child) edge of this type (1–7)
  • node_id — string — Node id e.g. "E34"; omit for the whole-graph summary
  • min_reach — integer — Filter to nodes with coordination_reach >= this value

  Hints: read-only yes · destructive no · idempotent yes · open-world no

• traverse_coordination read-only

  Walk the Coordination Topology lineage graph (precomputed lookups over the parent→child Coordination Connection DAG). `mode="path"` returns every path between `source` and `target`, each with its CC-type (connection-type) sequence and whether it is pure lineage (homogeneous) or crosses a sibling/bidirectional edge; `mode="ancestors"`/`"descendants"` return the nodes reachable up/down the hierarchy; `mode="neighbours"` returns direct parents and children. For a node's structural metrics or the whole-graph summary rather than walks, use `get_topology`. `cc_types` (connection-type filter) and `max_depth` apply to `path` AND to the `ancestors`/`descendants`/`neighbours` adjacency walks — supplying either runs a bounded typed walk over the backbone edges. `homogeneous_only` is `path`-only (drops any path crossing a sibling edge); sent to an adjacency mode it is returned in `ignored_params`. A bare numeric `node_id` missing the `E` prefix is resolved (the canonical form is echoed as `resolved_node_id`); a node that exists but carries no backbone edges is reported as such rather than as unknown.

  Parameters:

  • mode (required) — string — Traversal mode
  • source — string — Source node id for mode=path
  • target — string — Target node id for mode=path
  • node_id — string — Node id for ancestors | descendants | neighbours (e.g. "E93"; a bare "93" is resolved)
  • cc_types — array — Edge-type (connection-type) filter, applied to path and adjacency walks; default = all types
  • max_depth — integer — Maximum walk depth — path length in mode=path, hop count from node_id in the adjacency modes
  • homogeneous_only — boolean — mode=path only — true → drop any path crossing a sibling (Type 6/7) edge; ignored (and echoed in ignored_params) on adjacency modes

  Hints: read-only yes · destructive no · idempotent yes · open-world no

• get_capability_signatures read-only

  Retrieve the precomputed Capability Signatures — directional Coordination Connection (CC) chains or hubs (connection types 1/3/4/5) whose member events span two or more Sovereign Capability Profile (SCP) dimensions, surfaced as corroborating structural texture rather than a headline finding. Omit `signature_id` for the whole set — each signature with its nodes, directional legs (from/to + connection type), dimension span, attribution profile, and narrative; its `member_links`, the explicit list of Capability Link `link_id`s its nodes correspond to; and its own lifecycle fields — `created`, `scp_register_version`, `assessed_under`, `status` — plus metadata (the directional-edge inventory, the active/retired `status_counts`, and the pre-operational caveat) and the non-signature same-capability clusters. Pass `signature_id` (e.g. "CS-001") for one. A signature is never edited in place: a change to its member set retires it whole, and a still-qualifying structure re-enters as a new signature. Active signatures are returned by default; pass `include_retired` to receive retired signatures alongside. `member_links` is the authoritative reference into the membership layer: the node fields are denormalised for rendering and are subordinate to it, so resolve a `link_id` through `get_capability_links` for that node's current status and full detail. For the underlying event↔dimension membership links, use `get_capability_links`. A derived overlay over the directional CC graph; served in full and free, lifecycle fields included.

  Parameters:

  • signature_id — string — Signature id e.g. "CS-001"; omit for all signatures
  • include_retired — boolean — Include retired signatures alongside active ones (default false — active only). A retired signature is the frozen whole-state snapshot of a structure that changed or ceased to qualify; use this to reconstruct a past state.

  Hints: read-only yes · destructive no · idempotent yes · open-world no

• get_capability_links read-only

  Retrieve the event↔dimension membership links — which events qualify or contribute to each actor's Sovereign Capability Profile (SCP) capability dimensions. Filter by `actor` (e.g. "US", "CN", "GB" — note GB, not UK), `dimension` (D1–D7), or `event_id`. Each link carries its `link_type` (`qualifying_contributor` = establishes the dimension vs `capability_area` = contributes to it), `attribution` (specific | general), and `entity` (the named asset, on qualifying links only); its identity and lifecycle — `link_id` (the stable handle, CL-{ACTOR}-{Dn}-{eventID}.{sequence}), `created`, `scp_register_version` (the register cycle it was assessed in), `assessed_under` (the methodology version in force at that assessment, which may be older than the dataset's current one), `status`, and on retired links `status_changed`; and at full tier `qualification_basis`, `rationale` and `note` (the deep qualification analysis), `status_reason` (the assessed reasoning for a retirement), and `provenance` (construction-cohort marker). Sole-vs-multiple basis is derivable by counting `qualifying_contributor` links per actor×dimension (1 = sole; >1 = one of several) — links are never flattened to `qualifies:yes/no`. Active links are returned by default; retired links are never restored, and a relationship assessed afresh enters as a new link at the next sequence. Pass `include_retired` to receive the archive alongside the active links. The response carries a metadata block — totals by type, per-actor and per-dimension distributions, active/retired `status_counts`, met-dimension coverage figures, and the dataset's generation date and revision history — identical on both tiers; `last_updated` dates the most recent revision. For the higher-order signatures built from these links, use `get_capability_signatures`.

  Parameters:

  • actor — string — Actor code filter: a Sovereign Capability Profile register code (ISO 3166-1 alpha-2; EU for the European Union); `get_actors` lists the register. Named `actor_name` in `get_actors`.
  • event_id — string — Event id filter, e.g. "196"
  • dimension — string — Dimension code filter: D1–D7
  • include_retired — boolean — Include retired links alongside active ones (default false — active only). Retired links are frozen snapshots of links that no longer qualify; use this to reconstruct a past state, not to read current membership.

  Hints: read-only yes · destructive no · idempotent yes · open-world no

• get_threads read-only

  Retrieve Infrastructure Threads — the factual material-link layer recording where the same named AI hardware asset (chip / GPU / NPU / custom ASIC) appears as a documented fact in two events. A thread is not a Coordination Connection: no connection type, no confidence grade — either the asset match is verified or it isn't. Pass `event_id` for that event's linked events in both directions: outbound = assets this event deploys or trains on, each linking to its producer event; inbound = events that use this event's asset. Each link carries the sourced claim (`linkType` ∈ deploys | trains_on | powered_by | fabricated_at, plus the asset) and the host-event field that names it. Omit `event_id` for the whole layer. For one event's value-chain stack, use `get_event_stack`. Free on both tiers.

  Parameters:

  • event_id — string — Event id e.g. "182" (a deployer) or "196" (a producer); omit for the whole thread layer

  Hints: read-only yes · destructive no · idempotent yes · open-world no

• get_event_stack read-only

  Assemble the stack journey for one event — its kind-labelled value-chain neighbourhood at depth 1, the same composition the Atlas renders, in one call. Returns three separate arrays, never merged: `coordination` — the event's own Coordination Connections (CCs), all seven connection types, directed (types 1/3/4/5) and bidirectional/sibling (types 2/6/7) alike; each carries type/confidence/direction, plus evidence at full tier; `material` — the event's own Infrastructure-Thread edges, one per source→target pair, each with its sourced asset facts (`linkType` + asset + naming field); `capability` — the active Sovereign Capability Profile (SCP) dimension-membership waypoints for the event itself (actor, dimension, link_type, attribution, entity, and the link's `link_id` and lifecycle fields; plus qualification_basis/rationale/note at full tier). The three are different kinds of claim: a thread is a factual asset match, not a CC, and is never rendered as one. The capability array carries active links only and takes no parameter to widen it — a stack is the event's neighbourhood as the corpus now stands; for retired memberships, use `get_capability_links` with `include_retired`. A derived view over the existing connections, threads, and capability links — each substrate keeps its own gating, and the free tier leaks no paid fields. For the material layer alone, use `get_threads`; for the directed multi-hop coordination lineage between events, use `traverse_coordination`. Pass `event_id`.

  Parameters:

  • event_id (required) — string — Focal event id e.g. "196" — the event whose stack neighbourhood to assemble

  Hints: read-only yes · destructive no · idempotent yes · open-world no

• query_scp read-only

  Query the Sovereign Capability Profile (SCP) register by pivoting and comparing — distinct from `get_actors`, which fetches whole actors and is where `energy_context` is served; it is not returned here. Pass exactly one of: `dimension` (e.g. "D3") pivots that one dimension across every actor (each actor's met status and recorded headline, on what basis); `actors` (e.g. ["US","CN"]) lays those actors side by side across all seven dimensions (D1–D7); `watch:true` returns the `dimension_watch` register (forward-looking entries: actor, dimension, current vs expected_change, trigger, timeline). Readouts are descriptive — recorded status and fields, never causal inference. Free tier carries `met` + the headline for all actors and the US sample in full; the full tier adds the per-dimension evidence (constraint, qualification, sources, and `escalation_trigger` where recorded). Omit all three params for a usage hint.

  Parameters:

  • watch — boolean — true → return the dimension_watch register
  • actors — array — Actor ids e.g. ["US","CN"] → compare those actors across D1–D7
  • dimension — string — Dimension id D1–D7 (e.g. "D3") → pivot that dimension across all actors

  Hints: read-only yes · destructive no · idempotent yes · open-world no

• get_blocs read-only

  Retrieve the WEO Bloc Membership Register — every nation in the register with its bloc assignment and universal membership category (Core / Integrated / Engaged / Peripheral / Non-Aligned), its key evidence and source URLs, plus document metadata and sectioned provenance (snapshot history, changelog, pre-register decisions). Served whole and free — the machine mirror of the public blocs register. A descriptive roster, never inference; the category definitions themselves are served by `get_methodology`.

  Hints: read-only yes · destructive no · idempotent yes · open-world no

• get_contribution read-only

  Look up WEO Contribution Architecture vocabulary — the contributor programme's terms, credit classes, and governance provisions (e.g. "Delta Credit", "Founding Observer", "Observer Network", "rate card", "malinformation"). Returns the term's context, its section anchor, and a deep link into the self-hosted CA edition. Omit `term` for programme status: phase, activation criterion, current corpus size, and enquiry address. Use to resolve participation vocabulary — the Contribution Architecture governs participation, whilst the Methodology Manual (`get_methodology`) governs what qualifies. Matching is exact-first, then substring; an unknown term returns a sample of available terms. Served in full on both tiers.

  Parameters:

  • term — string — The CA term to look up (e.g., "Delta Credit", "Founding Observer", "membership facts"). Omit for programme overview.

  Hints: read-only yes · destructive no · idempotent yes · open-world no

• get_metrics read-only

  Retrieve the WEO quantitative coordination metrics — dated observations of the documented connection network computed under Towards Coordination Science (TCS). Returns Cross-Sector Transmission Frequency (CSTF) with its permutation null model, the cross-sector transmission matrix (sparse non-zero cells plus the fixed 11-category row order), Framework Family Density, Cross-Bloc Connection Frequency, and Connection Growth Rate. For live corpus counts and distributions, use `get_coverage_stats` — metrics here are derived, dated observations, never live recomputations. Omit `metric` for all of them; set `include_history` for the full append-only series rather than the latest observation. Pre-operational: the corpus has not reached operational threshold, so every value is a dated observation of the documented record as constructed, not a measurement of the real-world coordination landscape. Each metric carries its own `status` — `published` metrics are surfaced on the platform; `pre_operational` ones are recorded but not publication-grade; `insufficient_corpus` ones are blocked. Values are comparable only within a `taxonomy` epoch. Served in full on both tiers.

  Parameters:

  • metric — string — Single metric key: cstf, transmission_matrix, ffd, cbcf, or cgr. Omit for all metrics.
  • include_history — boolean — Return the full append-only observation series instead of the latest observation only (default false).

  Hints: read-only yes · destructive no · idempotent yes · open-world no

• get_commitments read-only

  Retrieve the Compute Commitment Register — verified commitments of computing capacity by named actors to societal challenges, one record per instrument, quantities as published, never summed. Returns the challenge table (documents counted once per challenge touched; Committed and Allocated counts), the actor table, records with commitment, delivery and corroborating evidence, the Challenge List Reference, and `definitions` (each register term's defining sentence from the published working note, deep-linked). The register touches events only through Challenge Attributions — for an attributed event use `get_event`; for sovereign capability use `query_scp`; for coordination metrics use `get_metrics`. Pass `view` as `table` (default), `actors`, `records` or `challenges`; filter with `challenge`, `actor_id`, `actor_type`, `band`, `status`, `ai_compute` or `standing`; `record_id` returns one record in full, line items included; `include_history` returns the append-only observation series. Figures are dated observations within a Challenge List epoch — floors of located disclosure, never measures of activity; status stays `pre_operational` until the register's stated conditions are met. Served in full on both tiers.

  Parameters:

  • band — string — Filter by band: A grant or credit programme; B allocation on a named system; C sovereign or multilateral infrastructure; D commercial or philanthropic supply.
  • view — string — `table` (default): the challenge table with the sub-threshold table, transparency line and concentration. `actors`: the actor table. `records`: every live record without line items. `challenges`: the Challenge List Reference.
  • status — string — Filter by record status.
  • actor_id — string — Filter to one actor of record by register actor id, e.g. "CCR-ACT-0001".
  • standing — string — Filter to records with at least one line item of that standing.
  • challenge — string — Filter to records with a line item under this challenge (case-insensitive substring; published or sub-threshold).
  • record_id — string — Return one record in full, line items included, e.g. "CCR-0004". Overrides `view`.
  • actor_type — string — Filter records by actor type (Register Standard vocabulary).
  • ai_compute — string — Filter by whether the instrument states the capacity is for AI.
  • include_history — boolean — Return every observation in each series instead of the latest only (default false).

  Hints: read-only yes · destructive no · idempotent yes · open-world no

Repository

Measured by MCPYP probing. Untested fields are omitted, never estimated. Listing is always free; MCPYP never proxies calls or stores user tokens.