Ephemeris Time-Series Forecasting

Measured conclusion: Alive, not deep-tested. L2 outcome: not_attempted.

Probabilistic time-series forecasts from zero-shot foundation models: routed, single or ensembled.

Add to your agent

Copy this into your MCP client config:

{
  "mcpServers": {
    "Ephemeris Time-Series Forecasting": {
      "url": "https://ephemeris.cascade.industries/api/mcp",
      "transport": "streamable-http"
    }
  }
}

Measured tools (7)

• get_version read-only

  Read this deployed MCP server's version, nullable build commit and supported saved forecast schema versions. Call with {}. No API key, account access, inference, charges or writes; network access to this server only. Returns component, version, build_commit (full source SHA or null when unavailable), artifact_schema_versions, and runtime_tls (Node release and certificate-trust status; metadata only, no probe connections). Unknown commit stays null. Server, CLI, npm bridge and artifact schema versions are independent; this tool cannot inspect local installations, check the latest registry release or establish model quality. Use list_models for live model capabilities. After a connection failure reconnect and retry; no forecast could have been submitted by this tool.

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

• describe_forecast read-only

  Deterministic calculations from a complete ephemeris.forecast v1 artifact object, including CLI artifacts. No inference or forecast charge; this hosted tool still receives data over the network. Returns scope, result, basis, limitations and source pointers: last-observed baseline, median endpoint/change, marginal-median aggregates, and exact saved billing units. Missing medians/context remain unknown. Sum of medians only for period_total; it is not a cumulative-distribution interval or median. No calibrated coverage, causal explanation, path/peak probability, or business action is inferred. Example: pass {artifact: <forecast result artifact>}; choose series_index/variate_index for multiple results. Maximum 1 MiB; malformed/version/shape errors are recoverable by fixing the artifact or using CLI, never rerunning inference. Missing labels may be corrected in a COPY of context without changing request/response. No filesystem access, URL fetches, persistent storage, Gnomon import, or scoring.

  Parameters:

  • artifact (required) — object — Complete ephemeris.forecast schema_version 1 JSON object returned by forecast or saved by Ephemeris CLI. Pass the object, not a filename, URL, resource URI, request ID, or bearer token. Maximum 1 MiB UTF-8 JSON. Contains submitted history; send only data you are authorized to process on this hosted server.
  • series_index — integer — Zero-based series index; required when the artifact has multiple series, otherwise 0.
  • variate_index — integer — Zero-based variate index; required for multivariate results, otherwise 0.

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

• plot_forecast read-only

  Render a complete ephemeris.forecast v1 artifact object as PNG (default inline preview) or SVG (embedded resource to save). No inference or forecast charge; hosted processing receives the artifact over the network. Example: {artifact: <saved artifact>, format: png}. Requires returned q0.5; never invents a median. Observed history, forecast boundary, median and nominal quantile bands are labeled. Default history view is most recent max(24, 2*horizon) points, capped by available history; history: all shows all. Dates only when supported by recorded metadata, otherwise sampling steps. Bands do not establish calibrated coverage; missing/crossed points are omitted and counted. Maximum 1 MiB input. Select series_index/variate_index for multiple results. Fix argument/shape errors without rerunning inference. User-supplied labels can be corrected in a COPY of artifact context offline; no new forecast needed. Client decides preview/download UI. URNs identify attached bytes, not hosted URLs; no filesystem access, URL fetching or persistent storage.

  Parameters:

  • artifact (required) — object — Complete ephemeris.forecast schema_version 1 JSON object returned by forecast or saved by Ephemeris CLI. Pass the object, not a filename, URL, resource URI, request ID, or bearer token. Maximum 1 MiB UTF-8 JSON. Contains submitted history; send only data you are authorized to process on this hosted server.
  • lower — number — Existing lower quantile. Supply together with upper. Default pairs complementary available levels.
  • upper — number — Existing upper quantile, greater than lower. Supply together with lower.
  • format — string — png returns an inline image preview; svg returns an embedded text resource to save as .svg. Neither creates a hosted download URL.
  • history — Observed points to show: default min(available, max(24, 2*horizon)). all shows full history. View changes never change the artifact or inference.
  • series_index — integer — Zero-based series index; required when the artifact has multiple series, otherwise 0.
  • variate_index — integer — Zero-based variate index; required for multivariate results, otherwise 0.

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

• forecast

  Submit numeric observations to forecast future sampling steps. Spends credits on successful calls. Choose and retain idempotency_key before submitting; no automatic retries, replay depends on server retention. Minimal example: {mode: route, series: [{values: [10,12,11], freq: D}], horizon: 2, quantiles: [0.1,0.5,0.9]}. Returns forecasts and meta as before, plus artifact by default for describe_forecast/plot_forecast. Save the embedded JSON attachment: it includes full input history and local context; no hosted storage is created. Context never becomes a model input. Billing is at meta.billing in millicredit strings (100 mc = 0.1 credits). Missing future inputs mean not supplied, not no events. Quantile bands do not establish coverage, cumulative intervals or peak probabilities. After uncertain completion, reuse the same key and API request; an expired key may cause a new charge.

  Parameters:

  • mode (required) — string — "route": automatically select a model. "ensemble": blend compatible models and may cost more. Neither guarantees accuracy or calibration. "explicit": run the single model named in `model`.
  • series (required) — array — One to 64 series forecast in one request. All share horizon and quantiles.
  • model — string — Required when mode is "explicit". Must be a name returned by list_models.
  • top_k — integer — Ensemble only: cap on how many models participate.
  • combine — string — Ensemble only. "mixture" averages the predictive distributions (default); "vincentize" averages the quantiles.
  • context — object — Local artifact metadata, never sent as model inputs. series aligns with input series; timestamps align with history and future_timestamps with horizon. Units/target/measurement/timezone must come from the user; unknowns stay null. Future-input labels are caller assertions; narrative assumptions do not change inference.
  • horizon — integer — Number of future steps to forecast, 1 to 4096. Defaults to 64. Some models stop short of this (max_horizon in list_models): route and ensemble skip them, explicit mode rejects the request.
  • quantiles — array — Up to 21 quantile levels strictly between 0 and 1, for example [0.1, 0.5, 0.9]. The response keys forecasts by these as decimal strings.
  • context_len — integer — Most recent points per variate to feed the model (history length does not affect pricing), 1 to 16384. Defaults to 256. Values beyond the model's context cap are truncated at the cap.
  • idempotency_key — string — Client-chosen key, 8 to 128 characters of letters, numbers, dot, underscore, colon or hyphen. A retry with the same key and identical API body replays while retained by the server. Retain the key before submission; expired keys may cause a new charge.
  • include_artifact — boolean — Default true: include a portable ephemeris.forecast v1 artifact and embedded JSON attachment (up to 1 MiB), with full input history. Save it client-side for describe_forecast/plot_forecast or the CLI. false returns the existing API response only. No persistent hosted storage is created.

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

• list_models read-only

  Discover the model panel: each model's availability right now (healthy), capabilities (multivariate, covariates, max_horizon, auto_max_horizon), its weight in an ensemble at each horizon, pinned and actually served weight revision, price in millicredits per series-slot per 64 forecast steps. History length has no billing multiplier. Call this before using forecast in explicit mode; the deployed panel can differ from any documentation.

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

• get_balance read-only

  Return the account's spendable credit balance in millicredits, excluding credits reserved by in-progress forecasts. 1000 millicredits equal one credit.

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

• get_usage read-only

  Return the account's forecast request log, newest first, with the mode, models used, series count, horizon, credits estimated and settled, HTTP status and latency of each request.

  Parameters:

  • limit — integer — Rows per page, 1 to 200. Defaults to 50.
  • offset — integer — Rows to skip. Use `pagination.next_offset` from the previous page.

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

Publisher website Repository

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