Anansi Data MCP

FREE. Resolves a partial name into exact model IDs and dataset names you can pass to the other tools; returns two string arrays (models, datasets), capped at 25 each. Use this FIRST whenever you have a fuzzy name like 'deepseek' or 'gas' and need the canonical identifier. Do NOT use it to retrieve values or search inside records — it matches names only; use anansi_price_current for prices or anans

FREE. Current per-token prices for LLMs as recorded at our last poll: usd_per_prompt_token and usd_per_completion_token (USD per single token, not per million), context window in tokens, and the observation timestamp (unix seconds). Use this to check what a model costs right now, or to compare a handful of models before routing. Do NOT use it to see how a price MOVED — it returns only the latest o

FREE. LLM price changes we observed in the last 7 days: per model, which field moved (prompt, completion, cache_read), the from/to values in USD per token, the percentage move, and the unix timestamp we detected it. Providers publish only their current catalogue and overwrite it silently, so this record exists nowhere upstream. Use it to detect that a model you depend on got cheaper, dearer, or va

PAID ($0.03 via x402). The COMPLETE observed price-change history for every tracked LLM, with no time window and no row cap: model, field, from/to in USD per token, percent move, and detection timestamp. Use it to audit a provider's pricing behaviour over the whole archive or to feed a cost model. Do NOT use it for a quick 'did anything change lately' check — anansi_price_changes_recent answers th

PAID ($0.05 via x402). Answers 'which model should I route to' in one call: returns the cheapest models satisfying hard constraints on price, context window and modality, ranked by prompt price ascending, PLUS every price change and deprecation affecting those candidates in the window. Use it when choosing or re-validating a routing target. Do NOT use it merely to look up one model's price (anansi

PAID ($0.005 via x402). Current state of every tracked key in one dataset, as recorded at our last poll, with provenance naming the source and licence. Use it for the newest snapshot of a whole dataset — prediction-market probabilities, gas by chain, perp funding, stablecoin pegs. Do NOT use it for LLM prices (anansi_price_current is free) and do NOT use it for change-over-time (anansi_data_change

PAID ($0.02 via x402). The full recorded time-series for ONE key in one dataset, oldest to newest, each row carrying its observation timestamp. Use it to chart or backtest a single series. Do NOT use it to compare many keys at once (anansi_data_latest returns them all for one moment) and do NOT expect history predating our first observation — the archive is days old and depth varies by dataset. In

PAID ($0.02 via x402). Field-level diffs across a whole dataset: for each key that changed, which field moved, its from and to values, and when. Use it to detect movement across many keys without pulling every series. Do NOT use it for LLM prices specifically — anansi_price_changes_recent is free and anansi_price_changes is the full version. Input: since is a unix timestamp; omit for the whole arc

PAID ($0.01 via x402). Full-text search across the CURRENT state of every sellable dataset at once, returning matching rows with their dataset name. Use it when you do not know which dataset holds a value. Do NOT use it to resolve a name to an identifier — anansi_search does that free — and note it searches current state only, not history. Input: q is a phrase match.

PAID ($0.03 via x402). Velocity ranking: which keys in a dataset moved fastest on one numeric field over a window, returned as rate of change with the underlying from/to values. Use it to find what is moving without reading every series. Do NOT use it to judge whether a move is unusual for that key — that is anansi_anomalies. Inputs: field must be a numeric field present in the dataset; window_s i

PAID ($0.05 via x402). Statistical outliers: keys whose latest move breaks their OWN historical baseline, returned with a z-score. Use it to separate a genuinely abnormal move from ordinary volatility. Do NOT use it to rank raw speed (anansi_movers) and treat results cautiously on a young dataset, where a short baseline inflates z-scores. Input: z_threshold defaults to 3.0.

PAID ($0.04 via x402). One-call composite for a single key: its latest value, its velocity, its rank against peers in the same dataset, and its anomaly z-score — the answer four separate calls would give. Use it when you have one specific key and want a verdict rather than raw series. Do NOT use it to survey many keys (anansi_movers) or to fetch a full series (anansi_data_history). Inputs: dataset